{"auto_groups":["gpt专用","default"],"data":[{"model_name":"claude-opus-4-7","description":"{\"zh\": \"Anthropic 上一代 Opus 旗舰 Claude Opus 4.7（2026-04 发布），首个引入 xhigh 推理档、并带来高分辨率视觉（长边 2576px）。100 万上下文，擅长最难的软件工程与智能体任务（已被 4.8 取代）。\", \"en\": \"Anthropic's previous Opus flagship, Claude Opus 4.7 (Apr 2026) — the first with the xhigh effort level and high-resolution vision (2576px). 1M context, strong on the hardest software-engineering and agentic tasks (superseded by 4.8).\"}","tags":"推理,编程,视觉,长上下文","vendor_id":3,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["claude kiro反代 0.2倍率","Claude 专用","default"],"official_price":{"input":5,"output":25,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"],"pricing_version":"5a90f2b86c08bd983a9a2e6d66c255f4eaef9c4bc934386d2b6ae84ef0ff1f1f"},{"model_name":"flux-2-pro","description":"{\"zh\": \"Black Forest Labs FLUX.2 系列图像生成/编辑模型,潜空间流匹配架构,最高约 4 百万像素输出,文字渲染、精确配色与角色一致性出色,支持最多 10 张参考图编辑。\", \"en\": \"Black Forest Labs FLUX.2 image generation/editing models with a latent flow-matching architecture, up to ~4-megapixel output, excellent text rendering, exact color matching and character consistency, and up to 10 reference images for editing.\"}","tags":"图像生成,文生图","vendor_id":13,"quota_type":1,"model_ratio":0,"model_price":0.05,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["image-generation","openai"]},{"model_name":"gpt-3.5-turbo","description":"{\"zh\": \"GPT-3.5 Turbo 系列,OpenAI 早期的低成本对话模型。能力已明显落后于现役模型,保留用于兼容老集成与成本极敏感的简单任务。带日期后缀的是固定快照版本。\", \"en\": \"The GPT-3.5 Turbo series, OpenAI's early low-cost chat model. It now lags well behind current models and is kept for compatibility with older integrations and for very cost-sensitive simple tasks. Date-suffixed names are pinned snapshots.\"}","tags":"高速,低成本,小型","vendor_id":2,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":3,"enable_groups":["default","svip"],"official_price":{"input":0.5,"output":1.5,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"kling-v2-6","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"视频生成","vendor_id":6,"quota_type":1,"model_ratio":0,"model_price":0.3,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.3349,"480p":0.3349,"720p":0.3349}},{"model_name":"whisper-1","description":"{\"zh\": \"OpenAI Whisper 语音转文字模型,支持多语种识别与翻译成英文,对口音和背景噪声较为稳健,是长期的通用 ASR 基线。\", \"en\": \"OpenAI's Whisper speech-to-text model, supporting multilingual transcription and translation into English. It is robust to accents and background noise and has long served as the general-purpose ASR baseline.\"}","tags":"语音识别","vendor_id":2,"quota_type":0,"model_ratio":15,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","audio"]},{"model_name":"grok-4.5","description":"{\"zh\": \"xAI Grok 4.5(2026 发布),xAI 面向编程、智能体与知识工作的推理模型(后继为 Grok 4.6),支持视觉输入、工具调用与实时联网,主打「Opus 级」能力而价格更低。\", \"en\": \"xAI Grok 4.5 (2026), xAI's reasoning model for coding, agentic tasks and knowledge work (succeeded by Grok 4.6) — with vision input, tool use and real-time web access, marketed as 'Opus-class' at a lower price.\"}","tags":"推理,编程,智能体,联网,视觉","vendor_id":10,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":3,"cache_ratio":0.25,"enable_groups":["default","grok专用"],"official_price":{"input":2,"output":6,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gemini-2.5-flash","description":"{\"zh\": \"Google Gemini 2.5 Flash,上一代高性价比主力,原生多模态、约 100 万 token 上下文,混合推理带可调「思考预算」,在低延迟高并发且仍需推理的任务上性价比最佳。\", \"en\": \"Google Gemini 2.5 Flash, the previous cost-effective workhorse — natively multimodal, ~1M-token context, hybrid reasoning with an adjustable thinking budget, best price-performance for low-latency, high-volume tasks that still need reasoning.\"}","tags":"推理,多模态,高性价比,长上下文","vendor_id":4,"quota_type":0,"model_ratio":0.15,"model_price":0,"owner_by":"","completion_ratio":8.3333,"cache_ratio":0.25,"enable_groups":["default"],"official_price":{"input":0.3,"output":2.5,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"o3-mini","description":"{\"zh\": \"OpenAI o3 推理系列,把算力花在回答前的思考上,擅长数学、科研与复杂多步问题。`-mini` 是低成本快速档,`-pro` 思考最久、准确率最高但最慢最贵。\", \"en\": \"The OpenAI o3 reasoning series spends compute thinking before answering, and excels at mathematics, research and complex multi-step problems. `-mini` is the cheap fast tier; `-pro` thinks longest for the highest accuracy but is the slowest and most expensive.\"}","tags":"推理,科研","vendor_id":2,"quota_type":0,"model_ratio":0.55,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.5,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"skyreels-v4-fast","description":"{\"zh\": \"SkyReels V4 视频生成模型,面向影视级短视频创作,含 Fast/Std 档。\", \"en\": \"SkyReels V4 video generation for cinematic short-form video, in Fast/Std tiers.\"}","tags":"视频生成","vendor_id":16,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"gpt-image-2","description":"{\"zh\": \"OpenAI 第三代旗舰图像模型（2026-04 发布），支持文生图与图像编辑，全新自回归架构、速度约为上代 3–5 倍，支持 1K/2K/4K 分辨率与多种比例、最多 16 张参考图，文字渲染近乎精准、多语言排版出色。\", \"en\": \"OpenAI's third-generation flagship image model (Apr 2026) for text-to-image and image editing, with a new autoregressive architecture ~3-5x faster than its predecessor, 1K/2K/4K resolutions, up to 16 reference images, near-perfect text rendering and strong multilingual layout.\"}","tags":"图像生成,图像编辑,文生图","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"gemini-2.5-flash-lite","description":"{\"zh\": \"Google Gemini 2.5 Flash-Lite(2025-07 GA),2.5 系列中最便宜、延迟最低、吞吐最高的一档,100 万 token 上下文,轻量推理并支持联网检索与代码执行,适合大规模低成本场景。\", \"en\": \"Google Gemini 2.5 Flash-Lite (GA Jul 2025), the cheapest, lowest-latency, highest-throughput tier of the 2.5 series — 1M-token context, lightweight reasoning with optional web-search grounding and code execution.\"}","tags":"多模态,高速,低成本,长上下文","vendor_id":4,"quota_type":0,"model_ratio":0.05,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.25,"enable_groups":["default"],"official_price":{"input":0.1,"output":0.4,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gpt-4-turbo","description":"{\"zh\": \"GPT-4 Turbo 系列,12.8 万 token 上下文、支持图像输入,较初代 GPT-4 更快更便宜。属于上一代模型,新项目建议用 GPT-4o 及以上。\", \"en\": \"The GPT-4 Turbo series: 128K-token context with image input, faster and cheaper than the original GPT-4. It is a previous-generation model - new projects should prefer GPT-4o or newer.\"}","tags":"长上下文,视觉","vendor_id":2,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":3,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4o-transcribe","description":"{\"zh\": \"基于 GPT-4o 的语音转文字模型,在准确率和对专有名词的还原上优于 Whisper,适合会议记录、字幕与语音输入。\", \"en\": \"A speech-to-text model built on GPT-4o, more accurate than Whisper and better at proper nouns, which suits meeting notes, subtitles and voice input.\"}","tags":"语音识别","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai","audio"]},{"model_name":"viduq2-pro","description":"{\"zh\": \"生数科技 Vidu 视频生成模型,支持文生视频、图生视频与多主体参考,生成快速。\", \"en\": \"Shengshu Vidu video generation for text-to-video and image-to-video with multi-subject reference and fast generation.\"}","tags":"视频生成","vendor_id":7,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"gemini-3.8-flash-tiered","description":"{\"zh\": \"Google Gemini 3.8 Flash(2026-09-02 发布,直接 GA),发布时被 Google 称为最聪明的 Flash 模型,面向编程、智能体、多模态推理与专业工作流,并针对长周期编程和自主智能体做了调优。100 万 token 输入、6.4 万输出,支持文本/图像/音频/视频/PDF 输入、函数调用、搜索工具与电脑操作;思考强度分 low/medium/high,默认 medium。`-high`/`-medium`/`-low` 后缀对应思考强度档位。\", \"en\": \"Google Gemini 3.8 Flash (September 2, 2026, launched directly as GA), billed at launch as Google's most intelligent Flash model, for coding, agents, multimodal reasoning and professional workflows, and tuned for long-horizon coding and autonomous agents. 1M-token input and 64K output, accepting text, image, audio, video and PDF, with function calling, search as a tool and computer use; thinking levels low/medium/high, medium by default. The `-high` / `-medium` / `-low` suffixes select the thinking level.\"}","tags":"多模态,长上下文,编程,智能体,高速","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"minimax-m2.5","description":"{\"zh\": \"MiniMax M2 系列开源大模型,面向编程与智能体任务。小数点后的版本号(2.5/2.7)代表迭代次序,数字越大越新;它们是 M3 的上一代。\", \"en\": \"MiniMax's open-source M2 series, aimed at coding and agentic tasks. The decimal version numbers (2.5 / 2.7) mark successive iterations, higher being newer; they precede M3.\"}","tags":"开源,编程,智能体","vendor_id":21,"quota_type":0,"model_ratio":1.05,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-sonnet-5","description":"{\"zh\": \"Anthropic 中端模型 Claude Sonnet 5（2026-06 发布，后继为 Sonnet 5.5），默认开启自适应思考、支持 xhigh、高分辨率视觉。能以 Sonnet 的价格达到接近 Opus 4.8 的编程与智能体质量——会规划、用工具（浏览器/终端）、自主运行并自我校验，是大规模跑智能体的高性价比之选。\", \"en\": \"Anthropic's mid-tier model Claude Sonnet 5 (Jun 2026; succeeded by Sonnet 5.5) — adaptive thinking on by default, xhigh support and high-res vision. Reaches near-Opus-4.8 coding and agentic quality at Sonnet cost — plans, uses tools (browser/terminal), runs autonomously and self-verifies. The cost-effective way to run agents at scale.\"}","tags":"推理,编程,智能体,均衡,高性价比","vendor_id":3,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"image_ratio":0.833333333333,"enable_groups":["Claude 专用","default","claude max","claude kiro反代 0.2倍率"],"official_price":{"input":2,"output":10,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gemini-3.7-flash-tiered","description":"{\"zh\": \"Google Gemini 3.7 Flash(2026-08 发布),是在 3.6 Flash 推理基础上的算法改进而非新的预训练模型,编程与智能体能力提升明显。100 万 token 输入、6.4 万输出,支持文本/图像/视频/音频/PDF 输入、函数调用、Google 搜索工具与电脑操作。`-high`/`-medium`/`-low` 后缀对应不同的思考强度档位。\", \"en\": \"Google Gemini 3.7 Flash (Aug 2026) is an algorithmic improvement on 3.6 Flash's reasoning rather than a new pretrained model, with a clear jump in coding and agentic ability. 1M-token input and 64K output, accepting text, image, video, audio and PDF, with function calling, Google Search as a tool and computer use. The `-high` / `-medium` / `-low` suffixes select thinking effort.\"}","tags":"多模态,长上下文,编程,智能体,高速","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-opus-4-5-20251101","description":"{\"zh\": \"Claude Opus 4.5 的固定日期快照(2025-11-01)。指向某一次确定的模型版本,行为不会随滚动更新漂移,适合需要可复现结果的评测与生产固化场景。\", \"en\": \"A pinned dated snapshot of Claude Opus 4.5 (2025-11-01). It points at one fixed model build whose behaviour does not drift with rolling updates, which suits reproducible evaluations and pinned production deployments.\"}","tags":"推理,编程,智能体,长上下文","vendor_id":3,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","Claude 专用","claude kiro反代 0.2倍率"],"official_price":{"input":5,"output":25,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"claude-opus-5.5","description":"{\"zh\": \"Anthropic Claude Opus 5.5(2026-09-22 发布),面向长时间运行的智能体编程与知识工作,也是官方推荐多数场景首选的模型。100 万 token 上下文、单次最多输出 12.8 万 token,知识截止 2026-06;自适应思考始终开启、不能关闭,默认力度 medium,可按需调高。官方价比 Opus 5 低 20%,缓存读取只要输入价的 5%。不支持强制指定 tool_choice。(claude-opus-5.5 是 claude-opus-5-5 的点号写法,两者是同一个模型、同价。)\", \"en\": \"Anthropic Claude Opus 5.5 (released September 22, 2026), built for long-running agentic coding and knowledge work, and Anthropic's recommended starting point for most workloads. 1M-token context with up to 128K output per response and a June 2026 knowledge cutoff; adaptive thinking is always on and cannot be disabled, with medium default effort that you can raise as needed. List price is 20% below Opus 5, and cache reads cost only 5% of the input price. Forced tool_choice is not supported. (claude-opus-5.5 is the dotted alias of claude-opus-5-5: the same model at the same price.)\"}","tags":"旗舰,推理,编程,智能体,长上下文","vendor_id":3,"quota_type":0,"model_ratio":2,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.05,"create_cache_ratio":1.25,"enable_groups":["claude max"],"official_price":{"input":4,"output":20,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-6.1-sol","description":"{\"zh\": \"OpenAI GPT-6.1 Sol(2026-09-29 发布),官方定位为以低于 GPT-6 Astra 的成本处理复杂编程与专业工作。约 105 万 token 上下文、最多 12.8 万 token 输出,知识截止 2026-04;可走 Responses 与 Chat Completions 接口,并在 Responses 中支持把任务分派给子智能体的多智能体模式(beta)。\", \"en\": \"OpenAI GPT-6.1 Sol (released September 29, 2026), positioned by OpenAI for complex coding and professional work at a lower cost than GPT-6 Astra. ~1.05M-token context, up to 128K output and an April 2026 knowledge cutoff; available on the Responses and Chat Completions APIs, with a multi-agent mode on Responses (beta) that delegates work to subagents.\"}","tags":"推理,编程,智能体,长上下文,高性价比","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"enable_groups":["gpt专用","default"],"supported_endpoint_types":["openai"]},{"model_name":"davinci-002","description":"{\"zh\": \"OpenAI 的传统补全(completion)基座模型,比 babbage-002 更大。同样不是对话模型,仅为兼容微调与遗留脚本保留。\", \"en\": \"A legacy OpenAI base completion model, larger than babbage-002. Also not a chat model, kept only for compatibility with fine-tuning and legacy scripts.\"}","tags":"小型","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4.1","description":"{\"zh\": \"OpenAI GPT-4.1（2025-04 发布），非推理型对话/指令模型，专长编程与指令跟随，原生 100 万 token 超长上下文、支持图像输入，是稳定可靠的编码主力。\", \"en\": \"OpenAI GPT-4.1 (Apr 2025), a non-reasoning chat/instruct model specialized in coding and instruction-following, with a native 1M-token context and image input — a reliable coding workhorse.\"}","tags":"编程,多模态,长上下文,指令跟随","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.25,"enable_groups":["default"],"official_price":{"input":2,"output":8,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"glm-5","description":"{\"zh\": \"智谱 GLM-5(2026-02 发布),GLM-5 系列的首个版本,稀疏 MoE 架构、百万级上下文,面向长程编码与智能体任务。开源权重。\", \"en\": \"Zhipu GLM-5 (Feb 2026), the first release of the GLM-5 series: a sparse MoE with million-token context aimed at long-horizon coding and agentic tasks. Open weights.\"}","tags":"开源,推理,编程,长上下文","vendor_id":5,"quota_type":0,"model_ratio":2,"model_price":0,"owner_by":"","completion_ratio":4.5,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"union-alpha","quota_type":0,"model_ratio":0.005,"model_price":0,"owner_by":"","completion_ratio":1,"cache_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"kimi-k3","description":"{\"zh\": \"月之暗面 Kimi K3(2026-07 发布,同月开放权重),原生多模态智能体旗舰:2.8 万亿总参、每 token 激活约 1040 亿的MoE(Stable LatentMoE,896 个专家里选 16 个 + 2 个共享专家),约 105 万 token 上下文,支持文本 + 图像 + 视频输入。适合带截图、设计稿与演示视频的复杂工程场景。\", \"en\": \"Moonshot Kimi K3 (Jul 2026, weights opened the same month), a natively multimodal agentic flagship: a 2.8T-parameter MoE activating about 104B per token (Stable LatentMoE routes each token to 16 of 896 experts plus 2 shared), with roughly 1.05M-token context and text, image and video input. It suits complex engineering work involving screenshots, design mockups and demo videos.\"}","tags":"旗舰,开源,推理,编程,智能体,多模态,长上下文","vendor_id":11,"quota_type":0,"model_ratio":10,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"enable_groups":["default","kimi专用","svip"],"official_price":{"input":20,"output":100,"currency":"CNY"},"supported_endpoint_types":["openai"]},{"model_name":"claude-fable-5","description":"{\"zh\": \"Anthropic 位于 Opus 之上的前沿模型 Claude Fable 5（2026-06 发布，后继为 Fable 5.1），首个公开的「Mythos 级」前沿模型，发布时在软件工程、知识工作、视觉与科学研究等几乎所有能力基准上达到最先进水平。100 万上下文，思考始终开启，面向最苛刻的推理与长程智能体任务。\", \"en\": \"Anthropic Claude Fable 5 (Jun 2026; succeeded by Fable 5.1), the frontier tier above Opus — the first publicly available 'Mythos-class' frontier model, state-of-the-art at launch across nearly all benchmarks (software engineering, knowledge work, vision, scientific research). 1M context, thinking always on, for the most demanding reasoning and long-horizon agentic work.\"}","tags":"前沿,推理,编程,科研,顶配","vendor_id":3,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Claude 专用","default","claude max","claude kiro反代 0.2倍率"],"official_price":{"input":10,"output":50,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"seedance-2.0","description":"{\"zh\": \"字节跳动 Doubao Seedance 视频生成模型,支持文生视频/图生视频,2.0 起支持约 2K 分辨率、更长片段与原生同步音频、多素材输入。\", \"en\": \"ByteDance Doubao Seedance video generation, supporting text-to-video and image-to-video; 2.0+ adds ~2K resolution, longer clips, native synchronized audio and multi-input.\"}","tags":"视频生成","vendor_id":9,"quota_type":1,"model_ratio":0,"model_price":0.5082,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":3.4,"480p":0.6975,"720p":1.5}},{"model_name":"text-embedding-3-large","description":"{\"zh\": \"OpenAI 文本嵌入模型,把文本编码为向量用于语义检索、相似度、聚类与 RAG;3-large 精度最高,3-small 更具性价比,ada-002 为经典版。\", \"en\": \"OpenAI text-embedding models that encode text into vectors for semantic search, similarity, clustering and RAG; 3-large is highest quality, 3-small is more cost-effective, ada-002 is the classic version.\"}","tags":"文本嵌入,向量检索","vendor_id":2,"quota_type":0,"model_ratio":0.4,"model_price":0,"owner_by":"","completion_ratio":0,"enable_groups":["svip","default"],"supported_endpoint_types":["openai","embeddings"]},{"model_name":"gemini-3-flash","description":"{\"zh\": \"Google Gemini 3 Flash,Gemini 3 世代的速度档,100 万 token 上下文、原生多模态输入,适合高并发与低延迟场景。\", \"en\": \"Google Gemini 3 Flash, the speed tier of the Gemini 3 generation, with a 1M-token context and native multimodal input, suited to high-concurrency, low-latency workloads.\"}","tags":"多模态,长上下文,高速","vendor_id":4,"quota_type":0,"model_ratio":0.05,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.1-pro-low","description":"{\"zh\": \"Google Gemini 3.1 Pro,Pro 档是 Gemini 的高能力档,长上下文与多模态理解突出。`-high`/`-low` 后缀对应不同的思考强度:high 更准但更慢更贵,low 反之。\", \"en\": \"Google Gemini 3.1 Pro. The Pro tier is Gemini's high-capability option, strong at long context and multimodal understanding. The `-high` / `-low` suffixes select thinking effort: high is more accurate but slower and dearer, low the reverse.\"}","tags":"推理,多模态,长上下文,旗舰","vendor_id":4,"quota_type":0,"model_ratio":0.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"official_price":{"input":2,"output":12,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.1-pro-preview","description":"{\"zh\": \"Google Gemini 3.1 Pro(2026),Gemini 3 Pro 的升级旗舰,原生多模态、100 万 token 上下文,广博的世界知识与跨模态高级推理,面向最复杂的推理与编程任务。\", \"en\": \"Google Gemini 3.1 Pro (2026), the successor flagship to Gemini 3 Pro — natively multimodal, 1M-token context, broad world knowledge and advanced cross-modal reasoning for the most complex reasoning and coding tasks.\"}","tags":"推理,多模态,编程,长上下文","vendor_id":4,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":2,"output":12,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-embedding-2-preview","description":"{\"zh\": \"Google Gemini 文本嵌入模型,将文本转为高质量多语言语义向量,面向语义检索、聚类与 RAG 检索增强。\", \"en\": \"Google Gemini text-embedding models that turn text into high-quality multilingual semantic vectors for retrieval, clustering and RAG.\"}","tags":"文本嵌入,向量检索","vendor_id":4,"quota_type":0,"model_ratio":0.05,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai","embeddings"]},{"model_name":"viduq3-turbo","description":"{\"zh\": \"生数科技 Vidu 视频生成模型,支持文生视频、图生视频与多主体参考,生成快速。\", \"en\": \"Shengshu Vidu video generation for text-to-video and image-to-video with multi-subject reference and fast generation.\"}","tags":"视频生成","vendor_id":7,"quota_type":0,"model_ratio":0.375,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.5096,"480p":0.007596,"540p":0.2912,"720p":0.4368}},{"model_name":"seedance-2.0-fast-face","description":"{\"zh\": \"字节跳动 Doubao Seedance 视频生成模型,支持文生视频/图生视频,2.0 起支持约 2K 分辨率、更长片段与原生同步音频、多素材输入。\", \"en\": \"ByteDance Doubao Seedance video generation, supporting text-to-video and image-to-video; 2.0+ adds ~2K resolution, longer clips, native synchronized audio and multi-input.\"}","tags":"视频生成","vendor_id":9,"quota_type":0,"model_ratio":52,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":5.09652,"480p":0.728,"720p":1.5652}},{"model_name":"seedream-5-0-pro","description":"{\"zh\": \"字节 Seedream 5.0 的 Pro 档(2026-02 发布),对标顶级图像模型:支持 2K 直出与 4K 增强,首次支持联网检索生图,能理解「静谧科技感」这类抽象提示词,并提供精确笔刷编辑。\", \"en\": \"The Pro tier of ByteDance Seedream 5.0 (Feb 2026), aimed at the top of the image-model field: 2K direct output with 4K enhancement, the first Seedream to support retrieval-augmented generation from the web, comprehension of abstract prompts, and precise brush editing.\"}","tags":"图像生成,文生图,图像编辑,联网","vendor_id":9,"quota_type":1,"model_ratio":0,"model_price":0.3276,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"],"image_pricing":{"1K":0.26643708,"2K":0.5329069200000001}},{"model_name":"gpt-5.6-terra","description":"{\"zh\": \"OpenAI GPT-5.6 系列的均衡档推理模型，面向日常专业工作，能力对标上一代 GPT-5.5 而价格约低一半。原生多模态、约 105 万 token 上下文，是性价比与质量兼顾的主力选择。\", \"en\": \"The balanced tier of OpenAI's GPT-5.6 series for everyday professional work, matching the prior GPT-5.5 at roughly half the price. Natively multimodal, ~1.05M-token context.\"}","tags":"推理,多模态,均衡","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.125,"enable_groups":["gpt专用","default"],"official_price":{"input":2,"output":12,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.8-flash-low","description":"{\"zh\": \"Google Gemini 3.8 Flash(2026-09-02 发布,直接 GA),发布时被 Google 称为最聪明的 Flash 模型,面向编程、智能体、多模态推理与专业工作流,并针对长周期编程和自主智能体做了调优。100 万 token 输入、6.4 万输出,支持文本/图像/音频/视频/PDF 输入、函数调用、搜索工具与电脑操作;思考强度分 low/medium/high,默认 medium。`-high`/`-medium`/`-low` 后缀对应思考强度档位。\", \"en\": \"Google Gemini 3.8 Flash (September 2, 2026, launched directly as GA), billed at launch as Google's most intelligent Flash model, for coding, agents, multimodal reasoning and professional workflows, and tuned for long-horizon coding and autonomous agents. 1M-token input and 64K output, accepting text, image, audio, video and PDF, with function calling, search as a tool and computer use; thinking levels low/medium/high, medium by default. The `-high` / `-medium` / `-low` suffixes select the thinking level.\"}","tags":"多模态,长上下文,编程,智能体,高速","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-2.5-pro","description":"{\"zh\": \"Google Gemini 2.5 Pro,上一代的高能力档,100 万 token 上下文、原生多模态,内置思考能力。已被 Gemini 3 系列取代,保留用于兼容老集成。\", \"en\": \"Google Gemini 2.5 Pro, the previous generation's high-capability tier with a 1M-token context, native multimodality and built-in thinking. Superseded by the Gemini 3 series and kept for compatibility.\"}","tags":"推理,多模态,长上下文","vendor_id":4,"quota_type":0,"model_ratio":0.625,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":1.25,"output":10,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gpt-5-chat","description":"{\"zh\": \"GPT-5 的对话档,非推理模型,响应快,面向常规问答与内容生成。`-latest` 后缀跟随 OpenAI 滚动更新,不固定快照。\", \"en\": \"The conversational tier of GPT-5: a non-reasoning model that answers fast, aimed at general Q\u0026A and content generation. The `-latest` suffix follows OpenAI's rolling updates rather than pinning a snapshot.\"}","tags":"均衡,高速","vendor_id":2,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.104,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.2","description":"{\"zh\": \"GPT-5.2 系列(2025-12 发布),提供 instant / thinking / Pro 三种模式。`-chat` 后缀为非推理的对话档;`-chat-latest` 跟随滚动更新,不固定快照。\", \"en\": \"The GPT-5.2 series (Dec 2025), offered in instant, thinking and Pro modes. The `-chat` suffix is the non-reasoning conversational tier; `-chat-latest` follows rolling updates rather than pinning to a snapshot.\"}","tags":"推理,均衡","vendor_id":2,"quota_type":0,"model_ratio":0.875,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-image-2.5-flare-firefly","description":"{\"zh\": \"OpenAI GPT Image 2.5 Flare 的另一条供给线路(firefly 线路),模型与 gpt-image-2.5-flare 相同,面向日常快速出图,支持文生图与图像编辑。定价待定,暂未开放调用。\", \"en\": \"An alternative supply route (firefly) for OpenAI GPT Image 2.5 Flare: the same model as gpt-image-2.5-flare, for fast everyday generation, with text-to-image and image editing. Pricing is pending, so calls are not yet open.\"}","tags":"图像生成,图像编辑,文生图","vendor_id":2,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"],"image_tier_ratios":{"quality:high":9,"quality:max":36,"quality:medium":2.25,"quality:xhigh":16,"resolution:2k":2,"resolution:4k":3.4}},{"model_name":"gemini-3.7-flash-low","description":"{\"zh\": \"Google Gemini 3.7 Flash(2026-08 发布),是在 3.6 Flash 推理基础上的算法改进而非新的预训练模型,编程与智能体能力提升明显。100 万 token 输入、6.4 万输出,支持文本/图像/视频/音频/PDF 输入、函数调用、Google 搜索工具与电脑操作。`-high`/`-medium`/`-low` 后缀对应不同的思考强度档位。\", \"en\": \"Google Gemini 3.7 Flash (Aug 2026) is an algorithmic improvement on 3.6 Flash's reasoning rather than a new pretrained model, with a clear jump in coding and agentic ability. 1M-token input and 64K output, accepting text, image, video, audio and PDF, with function calling, Google Search as a tool and computer use. The `-high` / `-medium` / `-low` suffixes select thinking effort.\"}","tags":"多模态,长上下文,编程,智能体,高速","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4o-mini","description":"{\"zh\": \"OpenAI GPT-4o mini（2024-07 发布），高性价比小模型，支持文本+图像输入、128K 上下文，面向大规模、低成本的通用场景。\", \"en\": \"OpenAI GPT-4o mini (Jul 2024), a cost-efficient small model with text+image input and 128K context, for large-scale, low-cost general use.\"}","tags":"多模态,小型,高速,低成本","vendor_id":2,"quota_type":0,"model_ratio":0.075,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.533333,"enable_groups":["default"],"official_price":{"input":0.15,"output":0.6,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.4-pro","description":"{\"zh\": \"GPT-5.4 系列(2026-03 发布)的 Pro 档,推理时间最长、深度最高的一档,面向高难度数学、科研与复杂工程问题。响应慢、单价高,不适合日常对话。\", \"en\": \"The Pro tier of the GPT-5.4 series (Mar 2026): the longest-thinking, deepest-reasoning tier, aimed at hard mathematics, research and complex engineering problems. It is slow and expensive, and not meant for everyday chat.\"}","tags":"推理,顶配,科研","vendor_id":2,"quota_type":0,"model_ratio":15,"model_price":0,"owner_by":"","completion_ratio":6,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"mj_video","description":"{\"zh\": \"Midjourney 图像/视频生成,以强烈的艺术性与美学质量著称。\", \"en\": \"Midjourney image/video generation, renowned for its artistic quality and aesthetics.\"}","tags":"视频生成","vendor_id":14,"quota_type":1,"model_ratio":0,"model_price":0.8,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"o1-mini","description":"{\"zh\": \"OpenAI o1 推理系列,是 o 系列的第一代,通过更长的思考链提升数学与科研类问题的准确率。`-mini` 为低成本档,`-preview` 是早期预览版。已被 o3 / o4 系列取代,保留用于兼容。\", \"en\": \"The OpenAI o1 reasoning series, the first of the o-series, improving accuracy on mathematics and research problems through longer chains of thought. `-mini` is the low-cost tier and `-preview` the early preview. Superseded by the o3 / o4 series and kept for compatibility.\"}","tags":"推理,科研","vendor_id":2,"quota_type":0,"model_ratio":0.55,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.5,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"qwen3.5-plus","description":"{\"zh\": \"阿里通义 Qwen3.5-Plus(2026-02 发布),基于视觉与文本混合 token 预训练,在通用知识与推理类基准上表现突出。\", \"en\": \"Alibaba Qwen3.5-Plus (Feb 2026), pretrained on mixed vision and text tokens, with strong results on general knowledge and reasoning benchmarks.\"}","tags":"多模态,视觉,长上下文","vendor_id":12,"quota_type":0,"model_ratio":0.7,"model_price":0,"owner_by":"","completion_ratio":6,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"viduq3-pro","description":"{\"zh\": \"生数科技 Vidu 视频生成模型,支持文生视频、图生视频与多主体参考,生成快速。\", \"en\": \"Shengshu Vidu video generation for text-to-video and image-to-video with multi-subject reference and fast generation.\"}","tags":"视频生成","vendor_id":7,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":1.1648,"480p":0.030384,"540p":0.5096,"720p":1.092}},{"model_name":"deepseek-v4-flash","description":"{\"zh\": \"DeepSeek 已于 2026-09 退役 V4 Flash,并把这个旧名称转到新一代 V4.1 Flash,在本站调用该名称实际由 V4.1 Flash 提供:100 万 token 上下文、最多 38.4 万 token 输出,原生支持图片理解,支持思考与非思考两种模式、JSON 输出和工具调用。响应快、成本低,适合高并发、对成本敏感的推理与编程场景。新接入建议直接用 deepseek-flash。\", \"en\": \"DeepSeek retired V4 Flash in September 2026 and routes this legacy name to the newer V4.1 Flash, so calls under this name here are served by V4.1 Flash: 1M-token context and up to 384K output, with native image understanding, thinking and non-thinking modes, JSON output and tool calls. Fast and low-cost for high-concurrency, cost-sensitive reasoning and coding. For new integrations, use deepseek-flash directly.\"}","tags":"推理,编程,高速,低成本,视觉","vendor_id":1,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":12,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":2,"output":8,"currency":"CNY"},"supported_endpoint_types":["openai"]},{"model_name":"deepseek-v4-pro","description":"{\"zh\": \"DeepSeek V4 家族的旗舰模型（2026 发布），1.6T 总参 / 49B 激活的高效稀疏 MoE，100 万 token 上下文，采用压缩稀疏注意力等新架构大幅降低长上下文算力与显存开销。提供 非思考 / 深度思考High / 深度思考Max 三档,擅长前沿级长上下文推理与编程。开源（MIT）。\", \"en\": \"The flagship of the DeepSeek V4 family (2026), an efficient sparse MoE (1.6T total / 49B active) with 1M-token context and a new compressed-attention architecture that sharply cuts long-context compute and memory. Three modes (Non-think / Think High / Think Max); excels at frontier-scale long-context reasoning and coding. Open-source (MIT).\"}","tags":"推理,编程,长上下文,开源,高效","vendor_id":1,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":6.666666666667,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":9,"output":27,"currency":"CNY"},"supported_endpoint_types":["openai"]},{"model_name":"deepseek-flash","description":"{\"zh\": \"DeepSeek 官方 API 里当前 Flash 档的调用名,对应 V4.1 Flash(2026-09-10 发布)。100 万 token 上下文、最多 38.4 万 token 输出,原生支持图片理解,支持思考与非思考两种模式、JSON 输出和工具调用,兼容 OpenAI 与 Anthropic 两种接口格式。响应快、成本低,适合高并发、对成本敏感的推理与编程场景。\", \"en\": \"The model name for the current Flash tier on DeepSeek's official API, i.e. V4.1 Flash (released September 10, 2026). 1M-token context and up to 384K output, with native image understanding, thinking and non-thinking modes, JSON output and tool calls, and compatibility with both the OpenAI and Anthropic API formats. Fast and low-cost for high-concurrency, cost-sensitive reasoning and coding.\"}","tags":"推理,编程,高速,低成本,视觉","vendor_id":1,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"official_price":{"input":2,"output":8,"currency":"CNY"},"supported_endpoint_types":["openai"]},{"model_name":"firefly-gpt-image-2","description":"{\"zh\": \"OpenAI GPT Image 2 的另一条供给线路(firefly 线路),模型与 gpt-image-2 相同,支持文生图与图像编辑。\", \"en\": \"An alternative supply route (firefly) for OpenAI GPT Image 2: the same model as gpt-image-2, with text-to-image and image editing.\"}","tags":"图像生成,图像编辑,文生图","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"gpt-5.5","description":"{\"zh\": \"OpenAI 上一代前沿推理模型（2026-04 发布），面向最复杂的专业与智能体任务。原生文本+视觉、约 105 万 token 上下文、128K 最大输出，推理力度可在 none/low/medium/high/xhigh 间调节。\", \"en\": \"OpenAI's previous-generation frontier reasoning model (Apr 2026) for the most complex professional and agentic tasks. Native text+vision, ~1.05M-token context, 128K max output; reasoning effort from none to xhigh.\"}","tags":"推理,多模态,旗舰,智能体","vendor_id":2,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.12,"enable_groups":["gpt专用","default"],"official_price":{"input":5,"output":30,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.1-flash-lite","description":"{\"zh\": \"Google Gemini 3.1 Flash-Lite,Gemini 家族里最轻最便宜的一档,面向大批量、低延迟、成本敏感的简单任务。\", \"en\": \"Google Gemini 3.1 Flash-Lite, the lightest and cheapest tier of the Gemini family, aimed at high-volume, low-latency, cost-sensitive simple tasks.\"}","tags":"高速,低成本,小型","vendor_id":4,"quota_type":0,"model_ratio":0.125,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":0.25,"output":1.5,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.3-chat-latest","description":"{\"zh\": \"GPT-5.3 系列。`-chat` 后缀是面向对话的非推理档,响应快、成本低;`-chat-latest` 跟随 OpenAI 滚动更新到该系列最新的对话版本,不固定在某个快照。\", \"en\": \"The GPT-5.3 series. The `-chat` suffix is the non-reasoning conversational tier: fast and cheap; `-chat-latest` follows OpenAI's rolling updates to the newest chat build in the series rather than pinning to one snapshot.\"}","tags":"推理,编程,均衡","vendor_id":2,"quota_type":0,"model_ratio":0.875,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"seedance-2.5","description":"{\"zh\": \"字节跳动 Doubao Seedance 视频生成模型,支持文生视频/图生视频,2.0 起支持约 2K 分辨率、更长片段与原生同步音频、多素材输入。\", \"en\": \"ByteDance Doubao Seedance video generation, supporting text-to-video and image-to-video; 2.0+ adds ~2K resolution, longer clips, native synchronized audio and multi-input.\"}","tags":"视频生成","vendor_id":9,"quota_type":1,"model_ratio":0,"model_price":0.6607,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":4.42,"480p":0.9068,"720p":1.95}},{"model_name":"flux-kontext-pro","description":"{\"zh\": \"Black Forest Labs FLUX.2 系列图像生成/编辑模型,潜空间流匹配架构,最高约 4 百万像素输出,文字渲染、精确配色与角色一致性出色,支持最多 10 张参考图编辑。\", \"en\": \"Black Forest Labs FLUX.2 image generation/editing models with a latent flow-matching architecture, up to ~4-megapixel output, excellent text rendering, exact color matching and character consistency, and up to 10 reference images for editing.\"}","tags":"图像生成,文生图","vendor_id":13,"quota_type":1,"model_ratio":0,"model_price":0.04,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["image-generation","openai"]},{"model_name":"happyhorse-1.0","description":"{\"zh\": \"阿里 ATH(HappyHorse)1.0 视频生成模型,2026-04 发布,支持 720p 与 1080p 输出,在综合评测榜上排名靠前。\", \"en\": \"Alibaba's ATH (HappyHorse) 1.0 video generation model, released April 2026, supporting 720p and 1080p output and ranking near the top of composite leaderboards.\"}","tags":"视频生成,文生视频","vendor_id":12,"quota_type":1,"model_ratio":0,"model_price":2.093,"owner_by":"","completion_ratio":0,"enable_groups":["svip","default"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":2.093,"480p":2.093,"720p":1.183}},{"model_name":"kimi-k2-thinking","description":"{\"zh\": \"月之暗面 Kimi K2 系列开源大模型,以 MoE 架构与长上下文见长,擅长编程与智能体任务。`-instruct` 为指令对话版,`-thinking` 为开启深度思考的版本,小数点后的版本号(2.5/2.6/2.7)代表迭代次序,数字越大越新。\", \"en\": \"Moonshot's open-source Kimi K2 series, built on MoE with long context and strong at coding and agentic tasks. `-instruct` is the instruction-tuned chat build and `-thinking` enables deep reasoning; the decimal version numbers (2.5 / 2.6 / 2.7) mark successive iterations, higher being newer.\"}","tags":"开源,编程,智能体,长上下文","vendor_id":11,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai"]},{"model_name":"qwen-image-2.0","description":"{\"zh\": \"阿里巴巴 Qwen-Image 2.0 统一图像生成+编辑模型,原生 2K 输出,中英双语排版/海报/信息图出色,蒸馏为 4 步快速生成,开源。\", \"en\": \"Alibaba Qwen-Image 2.0, a unified image generation and editing model with native 2K output, excellent Chinese/English typography, posters and infographics, distilled to a 4-step fast path; open-source.\"}","tags":"图像生成,文生图","vendor_id":12,"quota_type":1,"model_ratio":0,"model_price":0.182,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"z-image-turbo","description":"{\"zh\": \"阿里达摩院 Z-Image-Turbo,基于 Diffusion Transformer 的轻量文生图模型,约 60 亿参数,主打「9 步出图」的极速生成,在消费级硬件上也能本地跑,开源。\", \"en\": \"Alibaba DAMO's Z-Image-Turbo, a lightweight Diffusion-Transformer text-to-image model of about 6B parameters built around nine-step generation for speed. It runs locally on consumer hardware and is open source.\"}","tags":"图像生成,文生图,高速,开源","vendor_id":12,"quota_type":1,"model_ratio":0,"model_price":0.02,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"minimax-h3-max","description":"{\"zh\": \"fal 在 MiniMax H3 基础上后训练的变体,提示词遵循度与画面质感更强,并与其推理栈协同优化以提高吞吐。只支持图生视频与参考生视频,没有纯文生视频 —— 既不给图也不给参考素材的请求会被直接拒绝。分辨率仅 480P/768P,时长 5–15 秒,同样自带原生音轨,按输出秒数计费。\", \"en\": \"fal's post-trained variant of MiniMax H3, tuned for stronger prompt adherence and better aesthetics, and co-optimised with fal's inference stack for higher throughput. Image-to-video and reference-to-video only — there is no text-to-video endpoint, so a request carrying neither an image nor references is rejected outright. 480P and 768P only, 5–15 seconds, native audio track, billed per second of output.\"}","tags":"视频生成,图生视频,多模态","vendor_id":21,"quota_type":1,"model_ratio":0,"model_price":0.91,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"480p":0.1138,"768p":0.182}},{"model_name":"gpt-4-turbo-preview","description":"{\"zh\": \"GPT-4 Turbo 系列,12.8 万 token 上下文、支持图像输入,较初代 GPT-4 更快更便宜。属于上一代模型,新项目建议用 GPT-4o 及以上。\", \"en\": \"The GPT-4 Turbo series: 128K-token context with image input, faster and cheaper than the original GPT-4. It is a previous-generation model - new projects should prefer GPT-4o or newer.\"}","tags":"长上下文,视觉","vendor_id":2,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":3,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-haiku-4-5-20251001","description":"{\"zh\": \"Anthropic 最快、最便宜的模型 Claude Haiku 4.5（2025-10 发布），20 万上下文、64K 输出，以约 1/3 成本、两倍以上速度达到接近 Sonnet-4 级编程水平，适合延迟与成本敏感的对话、分类与廉价子任务。\", \"en\": \"Anthropic's fastest, cheapest model, Claude Haiku 4.5 (Oct 2025), 200K context and 64K output — near Sonnet-4-level coding at ~1/3 the cost and 2x+ the speed, for latency- and cost-sensitive chat, classification and cheap agent sub-tasks.\"}","tags":"高速,低成本,小型,视觉","vendor_id":3,"quota_type":0,"model_ratio":0.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Claude 专用","default","claude kiro反代 0.2倍率"],"official_price":{"input":1,"output":5,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"text-embedding-3-small","description":"{\"zh\": \"OpenAI 文本嵌入模型,把文本编码为向量用于语义检索、相似度、聚类与 RAG;3-large 精度最高,3-small 更具性价比,ada-002 为经典版。\", \"en\": \"OpenAI text-embedding models that encode text into vectors for semantic search, similarity, clustering and RAG; 3-large is highest quality, 3-small is more cost-effective, ada-002 is the classic version.\"}","tags":"文本嵌入,向量检索","vendor_id":2,"quota_type":0,"model_ratio":0.1,"model_price":0,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","embeddings"]},{"model_name":"veo3.1-fast","description":"{\"zh\": \"Google Veo 3.1 视频生成模型,支持文生视频/图生视频、高清带音频,含 Fast/Lite/Quality 多档。\", \"en\": \"Google Veo 3.1 video generation for text-to-video and image-to-video, high-definition with audio, in Fast/Lite/Quality tiers.\"}","tags":"视频生成","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":1.274,"owner_by":"","completion_ratio":0,"enable_groups":["svip","default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"claude-opus-5-5","description":"{\"zh\": \"Anthropic Claude Opus 5.5(2026-09-22 发布),面向长时间运行的智能体编程与知识工作,也是官方推荐多数场景首选的模型。100 万 token 上下文、单次最多输出 12.8 万 token,知识截止 2026-06;自适应思考始终开启、不能关闭,默认力度 medium,可按需调高。官方价比 Opus 5 低 20%,缓存读取只要输入价的 5%。不支持强制指定 tool_choice。\", \"en\": \"Anthropic Claude Opus 5.5 (released September 22, 2026), built for long-running agentic coding and knowledge work, and Anthropic's recommended starting point for most workloads. 1M-token context with up to 128K output per response and a June 2026 knowledge cutoff; adaptive thinking is always on and cannot be disabled, with medium default effort that you can raise as needed. List price is 20% below Opus 5, and cache reads cost only 5% of the input price. Forced tool_choice is not supported.\"}","tags":"旗舰,推理,编程,智能体,长上下文","vendor_id":3,"quota_type":0,"model_ratio":2,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.05,"create_cache_ratio":1.25,"enable_groups":["claude kiro反代 0.2倍率","claude max","default","Claude 专用"],"official_price":{"input":4,"output":20,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-4.1-nano","description":"{\"zh\": \"OpenAI GPT-4.1 家族中最快、最便宜的超轻量模型，100 万 token 上下文，适合对速度和成本极度敏感的高并发任务。\", \"en\": \"The fastest, cheapest ultra-light model in the GPT-4.1 family, 1M-token context, for latency- and cost-sensitive high-volume tasks.\"}","tags":"多模态,超轻量,高速,低成本","vendor_id":2,"quota_type":0,"model_ratio":0.05,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.3,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5-codex","description":"{\"zh\": \"GPT-5 系列的编程专用版本,针对代码生成、仓库理解与工具调用做过强化,配合 Codex 类客户端使用。\", \"en\": \"The coding-specialised build of the GPT-5 series, tuned for code generation, repository understanding and tool use, and designed to pair with Codex-style clients.\"}","tags":"编程,智能体","vendor_id":2,"quota_type":0,"model_ratio":0.625,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"kling-advanced-custom-elements","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"视频生成","vendor_id":6,"quota_type":0,"model_ratio":3.75,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"viduq2-turbo","description":"{\"zh\": \"生数科技 Vidu 视频生成模型,支持文生视频、图生视频与多主体参考,生成快速。\", \"en\": \"Shengshu Vidu video generation for text-to-video and image-to-video with multi-subject reference and fast generation.\"}","tags":"视频生成","vendor_id":7,"quota_type":0,"model_ratio":0.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"claude-opus-4-8","description":"{\"zh\": \"Anthropic Claude Opus 4.8（2026-05 发布，已被 Opus 5 / Opus 5.5 接替），100 万 token 上下文、128K 输出，擅长长程智能体执行、编程与知识工作。高度自主、可跑通宵级长任务并协调数百并行子智能体，比 4.7 更不易「默默放过」有缺陷的代码。自适应思考，力度可到 xhigh/max。\", \"en\": \"Anthropic Claude Opus 4.8 (May 2026; since succeeded by Opus 5 and Opus 5.5), 1M-token context and 128K output, excelling at long-horizon agentic execution, coding and knowledge work — highly autonomous, runs overnight-scale tasks with hundreds of parallel subagents, and far less likely than 4.7 to silently pass flawed code. Adaptive thinking up to xhigh/max.\"}","tags":"推理,编程,智能体,长上下文","vendor_id":3,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","claude kiro反代 0.2倍率","Claude 专用"],"official_price":{"input":5,"output":25,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"seedream-4-5","description":"{\"zh\": \"字节 Seedream 4.5(2025-12 发布),上一代图像生成模型,中文文字渲染与人像质感是它的强项。\", \"en\": \"ByteDance Seedream 4.5 (Dec 2025), the previous-generation image model, whose strengths are Chinese text rendering and portrait quality.\"}","tags":"图像生成,文生图,图像编辑","vendor_id":9,"quota_type":1,"model_ratio":0,"model_price":0.2075,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"grok-imagine-video","description":"{\"zh\": \"xAI Grok 正式视频模型(Grok Imagine / Aurora)。时长 1~15 秒任选,480P / 720P 可指定,成片自带同步音轨;支持文生视频与图生视频(参考图放 reference_images)。比 grok-video 全面更强。\", \"en\": \"xAI's proper Grok video model (Grok Imagine / Aurora). Any length from 1 to 15 seconds, selectable 480P / 720P, and a synchronized audio track; supports both text-to-video and image-to-video (reference images go in reference_images). A step up from grok-video in every way.\"}","tags":"视频生成,文生视频,图生视频,多模态","vendor_id":10,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"claude-fable-5-1","description":"{\"zh\": \"Anthropic Claude Fable 5.1(2026-09-01 发布),位于 Opus 之上的前沿档,面向最苛刻的推理与长周期智能体任务。相比 Fable 5,长时间智能体编程、多步研究,以及文档、表格、幻灯片类工作更强;输入输出价格不变,缓存读取降到原来的四分之一(仅为输入价的 2.5%)。100 万 token 上下文、单次最多输出 12.8 万 token,知识截止 2026-06;自适应思考始终开启,默认力度 high。不支持强制指定 tool_choice。\", \"en\": \"Anthropic Claude Fable 5.1 (released September 1, 2026), the frontier tier above Opus for the most demanding reasoning and long-horizon agentic work. Compared with Fable 5 it is stronger at long-running agentic coding, multistep research and document, spreadsheet and slide work, at the same input and output prices, with cache reads cut to a quarter (2.5% of the input price). 1M-token context with up to 128K output per response and a June 2026 knowledge cutoff; adaptive thinking is always on with high default effort. Forced tool_choice is not supported.\"}","tags":"前沿,推理,编程,科研,顶配","vendor_id":3,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.025,"create_cache_ratio":1.25,"enable_groups":["claude max"],"official_price":{"input":10,"output":50,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-5.2-chat","description":"{\"zh\": \"GPT-5.2 系列(2025-12 发布),提供 instant / thinking / Pro 三种模式。`-chat` 后缀为非推理的对话档;`-chat-latest` 跟随滚动更新,不固定快照。\", \"en\": \"The GPT-5.2 series (Dec 2025), offered in instant, thinking and Pro modes. The `-chat` suffix is the non-reasoning conversational tier; `-chat-latest` follows rolling updates rather than pinning to a snapshot.\"}","tags":"推理,均衡","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.4-mini","description":"{\"zh\": \"OpenAI GPT-5.4 家族的小型高速模型，发布时被官方称为「当时最强的小模型」，在编程、推理、多模态与工具调用上大幅提升，速度是上代 mini 的两倍多，兼顾能力与成本。\", \"en\": \"The small, fast model of the GPT-5.4 family — billed at launch as OpenAI's most capable small model, with big gains in coding, reasoning, multimodal and tool use and over 2x the speed of the previous mini.\"}","tags":"推理,多模态,小型,高速","vendor_id":2,"quota_type":0,"model_ratio":0.375,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":0.75,"output":4.5,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gpt-image-1.5","description":"{\"zh\": \"OpenAI GPT Image 系列图像生成与编辑模型,文字渲染精准、指令跟随强,支持多分辨率与参考图。\", \"en\": \"OpenAI GPT Image generation and editing models with precise text rendering, strong instruction-following, multiple resolutions and reference images.\"}","tags":"图像生成,文生图","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["image-generation","openai"]},{"model_name":"qwen3.7-plus","description":"{\"zh\": \"阿里通义 Qwen3.7-Plus(2026-06 发布),定位多模态一体化智能体基座,把「看、想、写、做、验」整合进同一套工作流,100 万 token 上下文,视觉理解能力在公开视觉榜单上位列全球前列。\", \"en\": \"Alibaba Qwen3.7-Plus (Jun 2026), positioned as an all-in-one multimodal agent base that folds seeing, thinking, writing, acting and verifying into one workflow, with a 1M-token context and vision understanding ranked among the best on public vision leaderboards.\"}","tags":"多模态,视觉,智能体,长上下文","vendor_id":12,"quota_type":0,"model_ratio":1.4,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"skyreels-v4-std","description":"{\"zh\": \"SkyReels V4 视频生成模型,面向影视级短视频创作,含 Fast/Std 档。\", \"en\": \"SkyReels V4 video generation for cinematic short-form video, in Fast/Std tiers.\"}","tags":"视频生成","vendor_id":16,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"nano-banana-pro","description":"{\"zh\": \"Google Nano Banana 系列的 Pro 档,相比标准档在细节、构图与指令遵循上更强,单张成本更高,适合成品图。\", \"en\": \"The Pro tier of Google's Nano Banana series, stronger than the standard tier on detail, composition and prompt adherence at a higher per-image cost - meant for finished images.\"}","tags":"图像生成,文生图,图像编辑","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":0.273,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"],"image_pricing":{"1K":0.273,"2K":0.273,"4K":0.3639909}},{"model_name":"gpt-6-sol","description":"{\"zh\": \"OpenAI GPT-6 Sol(2026-09 发布),官方定位为面向复杂编程与智能体工作流的模型,官方价是同代旗舰 GPT-6 Astra 的五分之一、上一代 GPT-5.6 Sol 的一半。支持文本+图像输入,约 105 万 token 上下文、12.8 万最大输出,推理强度 none 到 max 六档可调。内置工具与函数调用建议走 Responses API(Chat Completions 仅在 reasoning_effort=none 时支持函数调用)。\", \"en\": \"OpenAI GPT-6 Sol (released September 2026), positioned by OpenAI for complex coding and agentic workflows. Its list price is one-fifth of the flagship GPT-6 Astra and half of the previous GPT-5.6 Sol. Text and image input, ~1.05M-token context and 128K max output, with six reasoning-effort levels from none to max. Use the Responses API for built-in tools and function calling (Chat Completions supports function calling only with reasoning_effort=none).\"}","tags":"推理,编程,智能体,多模态,高性价比","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"enable_groups":["gpt专用","default"],"official_price":{"input":2,"output":10,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.1-flash-lite-preview","description":"{\"zh\": \"Google Gemini 3.1 Flash-Lite,Gemini 家族里最轻最便宜的一档,面向大批量、低延迟、成本敏感的简单任务。\", \"en\": \"Google Gemini 3.1 Flash-Lite, the lightest and cheapest tier of the Gemini family, aimed at high-volume, low-latency, cost-sensitive simple tasks.\"}","tags":"高速,低成本,小型","vendor_id":4,"quota_type":0,"model_ratio":0.125,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"glm-5.1","description":"{\"zh\": \"智谱 GLM-5.1(2026-04 发布),GLM-5 系列的第二个版本,长上下文与编程能力较 GLM-5 有提升。开源权重可商用。\", \"en\": \"Zhipu GLM-5.1 (Apr 2026), the second release in the GLM-5 series, improving on GLM-5 in long-context handling and coding. Open weights, available for commercial use.\"}","tags":"开源,推理,编程,长上下文","vendor_id":5,"quota_type":0,"model_ratio":3,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.22,"enable_groups":["default"],"official_price":{"input":6,"output":24,"currency":"CNY"},"supported_endpoint_types":["openai"]},{"model_name":"wan2.7-image","description":"{\"zh\": \"阿里巴巴通义万相(Wan)视频/图像生成模型,支持文生视频、图生视频与视频编辑。\", \"en\": \"Alibaba Tongyi Wan video/image generation, supporting text-to-video, image-to-video and video editing.\"}","tags":"图像生成,文生图","vendor_id":12,"quota_type":1,"model_ratio":0,"model_price":0.1966,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"suno","description":"{\"zh\": \"Suno 音乐生成模型:用一句话描述(灵感模式)或自定义歌词+风格(自定义模式)即可生成完整歌曲,支持带人声或纯器乐,附封面图与歌词。覆盖 v3.5–v5.5 多个版本。异步任务,通常 30–120 秒返回 2 首可选作品。\", \"en\": \"Suno music generation: turn a one-line prompt (inspiration mode) or custom lyrics + style (custom mode) into full songs — vocals or instrumental — complete with cover art and lyrics. Supports versions v3.5 through v5.5. Async task, typically 30–120s, returns 2 variations.\"}","tags":"音乐,歌曲生成,Suno,创作","quota_type":1,"model_ratio":0,"model_price":0.5,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai"]},{"model_name":"grok-4.6","description":"{\"zh\": \"xAI Grok 4.6(2026-08 发布,后继为 Grok 4.7),参数规模约 2 万亿,50 万 token 上下文,支持文本与图像输入,知识截止 2026-02。重点方向是长周期智能体、编程与知识工作;reasoning_effort 提供 low/medium/high/xhigh 四档。\", \"en\": \"xAI Grok 4.6 (Aug 2026; succeeded by Grok 4.7), around 2 trillion parameters with a 500K-token context and text plus image input, with a knowledge cutoff of February 2026. It focuses on long-horizon agents, coding and knowledge work, and offers four reasoning_effort levels: low, medium, high and xhigh.\"}","tags":"旗舰,推理,编程,智能体,长上下文","vendor_id":10,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":3,"cache_ratio":0.25,"enable_groups":["default","grok专用"],"official_price":{"input":2,"output":6,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gpt-image-2.5-sunburst-firefly","description":"{\"zh\": \"OpenAI GPT Image 2.5 Sunburst 的另一条供给线路(firefly 线路),模型与 gpt-image-2.5-sunburst 相同,支持文生图与图像编辑。定价待定,暂未开放调用。\", \"en\": \"An alternative supply route (firefly) for OpenAI GPT Image 2.5 Sunburst: the same model as gpt-image-2.5-sunburst, with text-to-image and image editing. Pricing is pending, so calls are not yet open.\"}","tags":"图像生成,图像编辑,文生图","vendor_id":2,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"],"image_tier_ratios":{"quality:high":9,"quality:max":36,"quality:medium":2.25,"quality:xhigh":16,"resolution:2k":2,"resolution:4k":3.4}},{"model_name":"gemini-3.8-flash-high","description":"{\"zh\": \"Google Gemini 3.8 Flash(2026-09-02 发布,直接 GA),发布时被 Google 称为最聪明的 Flash 模型,面向编程、智能体、多模态推理与专业工作流,并针对长周期编程和自主智能体做了调优。100 万 token 输入、6.4 万输出,支持文本/图像/音频/视频/PDF 输入、函数调用、搜索工具与电脑操作;思考强度分 low/medium/high,默认 medium。`-high`/`-medium`/`-low` 后缀对应思考强度档位。\", \"en\": \"Google Gemini 3.8 Flash (September 2, 2026, launched directly as GA), billed at launch as Google's most intelligent Flash model, for coding, agents, multimodal reasoning and professional workflows, and tuned for long-horizon coding and autonomous agents. 1M-token input and 64K output, accepting text, image, audio, video and PDF, with function calling, search as a tool and computer use; thinking levels low/medium/high, medium by default. The `-high` / `-medium` / `-low` suffixes select the thinking level.\"}","tags":"多模态,长上下文,编程,智能体,高速","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4.1-mini","description":"{\"zh\": \"GPT-4.1 系列的中间档,长上下文与指令跟随较上代明显改善,价格远低于旗舰,适合大批量的结构化处理与工具调用。\", \"en\": \"The mid tier of the GPT-4.1 series, with clearly better long-context handling and instruction following than the previous generation, at a price far below the flagship. Good for bulk structured processing and tool use.\"}","tags":"均衡,高速,低成本,长上下文","vendor_id":2,"quota_type":0,"model_ratio":0.2,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.25,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4o","description":"{\"zh\": \"OpenAI GPT-4o(omni),原生多模态模型,支持文本与图像输入,响应快于 GPT-4 Turbo 且更便宜,曾长期是 OpenAI 的通用主力。\", \"en\": \"OpenAI GPT-4o (omni), a natively multimodal model accepting text and image input. It is faster and cheaper than GPT-4 Turbo and was OpenAI's general-purpose workhorse for a long stretch.\"}","tags":"多模态,均衡,高速","vendor_id":2,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.5,"enable_groups":["default"],"official_price":{"input":2.5,"output":10,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.1-chat-latest","description":"{\"zh\": \"GPT-5.1 系列的对话档,并跟随 OpenAI 滚动更新到该系列最新版本。非推理模型,响应快,适合常规问答与内容生成。\", \"en\": \"The conversational tier of the GPT-5.1 series, following OpenAI's rolling updates to the newest build in that series. A non-reasoning model: fast, and suited to general Q\u0026A and content generation.\"}","tags":"均衡,高速","vendor_id":2,"quota_type":0,"model_ratio":0.625,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"claude-opus-5","description":"{\"zh\": \"Anthropic Claude Opus 5(2026-07 发布,后继为更便宜的 Opus 5.5),面向复杂智能体编码与企业工作。100 万 token 上下文、单次输出最高 12.8 万 token;思考默认开启,effort 从 low 到 max 五档可调。最大特点是自我验证 —— 会反复检查并迭代自己的产出直到通过,适合长时间自动化任务、调试与大型重构。\", \"en\": \"Anthropic Claude Opus 5 (Jul 2026; succeeded by the lower-priced Opus 5.5), built for complex agentic coding and enterprise work. 1M-token context with up to 128K output in one response; thinking is on by default across five effort levels from low to max. Its defining trait is self-verification - it re-checks and iterates on its own output until it passes - which suits long-running automation, debugging and large refactors.\"}","tags":"推理,编程,智能体,长上下文","vendor_id":3,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["claude kiro反代 0.2倍率","Claude 专用","default","claude max"],"official_price":{"input":5,"output":25,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"midjourney","description":"{\"zh\": \"Midjourney 图像/视频生成,以强烈的艺术性与美学质量著称。\", \"en\": \"Midjourney image/video generation, renowned for its artistic quality and aesthetics.\"}","vendor_id":14,"quota_type":1,"model_ratio":0,"model_price":0.4099,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4o-mini-transcribe","description":"{\"zh\": \"gpt-4o-transcribe 的轻量档,速度更快、单价更低,准确率略低,适合大批量或实时性要求高的转写。\", \"en\": \"The lightweight tier of gpt-4o-transcribe: faster and cheaper with slightly lower accuracy, suited to high-volume or latency-sensitive transcription.\"}","tags":"语音识别,低成本,高速","vendor_id":2,"quota_type":0,"model_ratio":0.15,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai","audio"]},{"model_name":"o1-preview","description":"{\"zh\": \"OpenAI o1 推理系列,是 o 系列的第一代,通过更长的思考链提升数学与科研类问题的准确率。`-mini` 为低成本档,`-preview` 是早期预览版。已被 o3 / o4 系列取代,保留用于兼容。\", \"en\": \"The OpenAI o1 reasoning series, the first of the o-series, improving accuracy on mathematics and research problems through longer chains of thought. `-mini` is the low-cost tier and `-preview` the early preview. Superseded by the o3 / o4 series and kept for compatibility.\"}","tags":"推理,科研","vendor_id":2,"quota_type":0,"model_ratio":7.5,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.5,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"minimax-m2.7","description":"{\"zh\": \"MiniMax M2 系列开源大模型,面向编程与智能体任务。小数点后的版本号(2.5/2.7)代表迭代次序,数字越大越新;它们是 M3 的上一代。\", \"en\": \"MiniMax's open-source M2 series, aimed at coding and agentic tasks. The decimal version numbers (2.5 / 2.7) mark successive iterations, higher being newer; they precede M3.\"}","tags":"开源,编程,智能体","vendor_id":21,"quota_type":0,"model_ratio":1.05,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"veo3.1-lite","description":"{\"zh\": \"Google Veo 3.1 视频生成模型,支持文生视频/图生视频、高清带音频,含 Fast/Lite/Quality 多档。\", \"en\": \"Google Veo 3.1 video generation for text-to-video and image-to-video, high-definition with audio, in Fast/Lite/Quality tiers.\"}","tags":"视频生成","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":0.637,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"gpt-6-astra","description":"{\"zh\": \"OpenAI GPT-6 Astra,新一代旗舰,官方定位为面向网络安全、专业工作、软件工程与科研的跨代升级。强调更快、更能守住任务边界、更擅长理解意图与完成多步工作流。\", \"en\": \"OpenAI GPT-6 Astra, the newest flagship generation, positioned by OpenAI as a generational step up for cybersecurity, professional work, software engineering and science. It emphasises speed, staying within task boundaries, understanding intent and completing multi-step workflows.\"}","tags":"旗舰,推理,编程,智能体,前沿","vendor_id":2,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.2,"enable_groups":["default","gpt专用"],"official_price":{"input":10,"output":50,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gpt-5","description":"{\"zh\": \"OpenAI GPT-5,把快速回答与深度推理合到一个模型里,按问题难度自动决定思考多久。综合能力强,是通用主力档。\", \"en\": \"OpenAI GPT-5 unifies fast answers and deep reasoning in a single model, deciding how long to think based on the difficulty of the question. A strong all-rounder and the general-purpose workhorse tier.\"}","tags":"推理,编程,均衡","vendor_id":2,"quota_type":0,"model_ratio":0.625,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5-chat-latest","description":"{\"zh\": \"GPT-5 的对话档,非推理模型,响应快,面向常规问答与内容生成。`-latest` 后缀跟随 OpenAI 滚动更新,不固定快照。\", \"en\": \"The conversational tier of GPT-5: a non-reasoning model that answers fast, aimed at general Q\u0026A and content generation. The `-latest` suffix follows OpenAI's rolling updates rather than pinning a snapshot.\"}","tags":"均衡,高速","vendor_id":2,"quota_type":0,"model_ratio":0.625,"model_price":0,"owner_by":"","completion_ratio":8,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5-nano","description":"{\"zh\": \"GPT-5 系列里最小最快的一档,单价最低,适合分类、抽取、改写、意图路由这类高频轻量任务。\", \"en\": \"The smallest and fastest tier of the GPT-5 series at the lowest price, suited to high-frequency lightweight tasks such as classification, extraction, rewriting and intent routing.\"}","tags":"高速,低成本,小型","vendor_id":2,"quota_type":0,"model_ratio":0.025,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"o3","description":"{\"zh\": \"OpenAI o3 推理系列,把算力花在回答前的思考上,擅长数学、科研与复杂多步问题。`-mini` 是低成本快速档,`-pro` 思考最久、准确率最高但最慢最贵。\", \"en\": \"The OpenAI o3 reasoning series spends compute thinking before answering, and excels at mathematics, research and complex multi-step problems. `-mini` is the cheap fast tier; `-pro` thinks longest for the highest accuracy but is the slowest and most expensive.\"}","tags":"推理,科研","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.25,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"flux-2-flex","description":"{\"zh\": \"Black Forest Labs FLUX.2 系列图像生成/编辑模型,潜空间流匹配架构,最高约 4 百万像素输出,文字渲染、精确配色与角色一致性出色,支持最多 10 张参考图编辑。\", \"en\": \"Black Forest Labs FLUX.2 image generation/editing models with a latent flow-matching architecture, up to ~4-megapixel output, excellent text rendering, exact color matching and character consistency, and up to 10 reference images for editing.\"}","tags":"图像生成,文生图","vendor_id":13,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["image-generation","openai"]},{"model_name":"kimi-k2.5","description":"{\"zh\": \"月之暗面 Kimi K2 系列开源大模型,以 MoE 架构与长上下文见长,擅长编程与智能体任务。`-instruct` 为指令对话版,`-thinking` 为开启深度思考的版本,小数点后的版本号(2.5/2.6/2.7)代表迭代次序,数字越大越新。\", \"en\": \"Moonshot's open-source Kimi K2 series, built on MoE with long context and strong at coding and agentic tasks. `-instruct` is the instruction-tuned chat build and `-thinking` enables deep reasoning; the decimal version numbers (2.5 / 2.6 / 2.7) mark successive iterations, higher being newer.\"}","tags":"开源,编程,智能体,长上下文","vendor_id":11,"quota_type":0,"model_ratio":2.1,"model_price":0,"owner_by":"","completion_ratio":5,"enable_groups":["default","svip"],"supported_endpoint_types":["openai"]},{"model_name":"veo3.1-fast-official","description":"{\"zh\": \"Google Veo 3.1 视频生成模型,支持文生视频/图生视频、高清带音频,含 Fast/Lite/Quality 多档。\", \"en\": \"Google Veo 3.1 video generation for text-to-video and image-to-video, high-definition with audio, in Fast/Lite/Quality tiers.\"}","tags":"视频生成","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.728,"480p":0.728,"4k":1.82,"720p":0.5824},"image_tier_ratios":{"quality:high":18,"quality:medium":4.5,"resolution:2k":2,"resolution:4k":3.4}},{"model_name":"wan2.6-i2v","description":"{\"zh\": \"阿里巴巴通义万相(Wan)视频/图像生成模型,支持文生视频、图生视频与视频编辑。\", \"en\": \"Alibaba Tongyi Wan video/image generation, supporting text-to-video, image-to-video and video editing.\"}","tags":"视频生成","vendor_id":12,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"wan2.7-r2v","description":"{\"zh\": \"阿里巴巴通义万相(Wan)视频/图像生成模型,支持文生视频、图生视频与视频编辑。\", \"en\": \"Alibaba Tongyi Wan video/image generation, supporting text-to-video, image-to-video and video editing.\"}","tags":"视频生成","vendor_id":12,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.9974,"480p":0.6042,"720p":0.6042}},{"model_name":"nano-banana-2","description":"{\"zh\": \"Google Nano Banana 2(即 Gemini 3.1 Flash Image,2026-02 发布),用 Flash 档的生成速度做到专业级画质,在光影、材质与整体美感上表现突出。注意:中文文字渲染是它的弱项。\", \"en\": \"Google Nano Banana 2 (Gemini 3.1 Flash Image, Feb 2026) delivers professional-grade image quality at Flash-tier speed, and stands out on lighting, texture and overall aesthetics. Note that rendering Chinese text is its weak point.\"}","tags":"图像生成,文生图,图像编辑,高速","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"gemini-3.8-flash-medium","description":"{\"zh\": \"Google Gemini 3.8 Flash(2026-09-02 发布,直接 GA),发布时被 Google 称为最聪明的 Flash 模型,面向编程、智能体、多模态推理与专业工作流,并针对长周期编程和自主智能体做了调优。100 万 token 输入、6.4 万输出,支持文本/图像/音频/视频/PDF 输入、函数调用、搜索工具与电脑操作;思考强度分 low/medium/high,默认 medium。`-high`/`-medium`/`-low` 后缀对应思考强度档位。\", \"en\": \"Google Gemini 3.8 Flash (September 2, 2026, launched directly as GA), billed at launch as Google's most intelligent Flash model, for coding, agents, multimodal reasoning and professional workflows, and tuned for long-horizon coding and autonomous agents. 1M-token input and 64K output, accepting text, image, audio, video and PDF, with function calling, search as a tool and computer use; thinking levels low/medium/high, medium by default. The `-high` / `-medium` / `-low` suffixes select the thinking level.\"}","tags":"多模态,长上下文,编程,智能体,高速","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3-pro-preview","description":"{\"zh\": \"Google Gemini 3 Pro（2025-11 发布),当时最先进的多模态推理旗舰,支持文本/图像/视频/音频/PDF 输入、100 万 token 上下文,具备动态思考与 Deep Think 深度思考模式,擅长复杂推理与智能体编码。\", \"en\": \"Google Gemini 3 Pro (Nov 2025), a leading multimodal reasoning flagship supporting text/image/video/audio/PDF input, 1M-token context, dynamic thinking plus a Deep Think mode — strong on complex reasoning and agentic coding.\"}","tags":"推理,多模态,编程,智能体,长上下文","vendor_id":4,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"seedance-2.0-fast","description":"{\"zh\": \"字节跳动 Doubao Seedance 视频生成模型,支持文生视频/图生视频,2.0 起支持约 2K 分辨率、更长片段与原生同步音频、多素材输入。\", \"en\": \"ByteDance Doubao Seedance video generation, supporting text-to-video and image-to-video; 2.0+ adds ~2K resolution, longer clips, native synchronized audio and multi-input.\"}","tags":"视频生成","vendor_id":9,"quota_type":1,"model_ratio":0,"model_price":0.2436,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":2.7,"480p":0.558,"720p":1.2}},{"model_name":"kling-v3-motion-control","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"视频生成","vendor_id":6,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.9362,"480p":0.9362,"720p":0.9362}},{"model_name":"gemini-3.1-flash-image-preview","description":"{\"zh\": \"Google Nano Banana(Gemini 图像模型),支持图像生成与多轮对话式编辑,文字渲染清晰、主体一致性强;Pro 版最高 4K、可融合多图并结合世界知识。\", \"en\": \"Google Nano Banana (Gemini image models) for image generation and multi-turn conversational editing, with clear text rendering and strong subject consistency; the Pro version reaches 4K and blends multiple images with world knowledge.\"}","tags":"图像生成,文生图","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"seedream-4-0","description":"{\"zh\": \"字节 Seedream 4.0 图像生成模型,支持文生图与图像编辑。属于较早的版本,新项目建议用 4.5 或 5.0。\", \"en\": \"ByteDance Seedream 4.0 for text-to-image and image editing. An earlier release - new projects should prefer 4.5 or 5.0.\"}","tags":"图像生成,文生图","vendor_id":9,"quota_type":1,"model_ratio":0,"model_price":0.1656,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"grok-video","description":"{\"zh\": \"xAI 的视频生成能力,基于 Grok Imagine / Aurora,支持文生视频与图生视频并可生成同步音效,面向快速创作短视频的场景。\", \"en\": \"xAI's video generation, built on Grok Imagine / Aurora — text-to-video and image-to-video with synchronized audio, for fast short-video creation.\"}","tags":"视频生成,文生视频,图生视频","vendor_id":10,"quota_type":1,"model_ratio":0,"model_price":0.03,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"gemini-2.5-flash-image","description":"{\"zh\": \"Google Nano Banana(Gemini 图像模型),支持图像生成与多轮对话式编辑,文字渲染清晰、主体一致性强;Pro 版最高 4K、可融合多图并结合世界知识。\", \"en\": \"Google Nano Banana (Gemini image models) for image generation and multi-turn conversational editing, with clear text rendering and strong subject consistency; the Pro version reaches 4K and blends multiple images with world knowledge.\"}","tags":"图像生成,文生图","vendor_id":4,"quota_type":0,"model_ratio":0.15,"model_price":0,"owner_by":"","completion_ratio":100,"cache_ratio":0.25,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"kling-omni-image","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"图像生成,文生图","vendor_id":6,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"kling-video-extend","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"视频生成","vendor_id":6,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"seedance-2.0-mini","description":"{\"zh\": \"字节跳动 Doubao Seedance 视频生成模型,支持文生视频/图生视频,2.0 起支持约 2K 分辨率、更长片段与原生同步音频、多素材输入。\", \"en\": \"ByteDance Doubao Seedance video generation, supporting text-to-video and image-to-video; 2.0+ adds ~2K resolution, longer clips, native synchronized audio and multi-input.\"}","tags":"视频生成","vendor_id":9,"quota_type":1,"model_ratio":0,"model_price":0.1016,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.68,"480p":0.1395,"720p":0.3}},{"model_name":"wan2.7","description":"{\"zh\": \"阿里巴巴通义万相(Wan)视频/图像生成模型,支持文生视频、图生视频与视频编辑。\", \"en\": \"Alibaba Tongyi Wan video/image generation, supporting text-to-video, image-to-video and video editing.\"}","tags":"视频生成","vendor_id":12,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.9974,"480p":0.6042,"720p":0.6042}},{"model_name":"gemini-3-pro-image-preview","description":"{\"zh\": \"Google Nano Banana(Gemini 图像模型),支持图像生成与多轮对话式编辑,文字渲染清晰、主体一致性强;Pro 版最高 4K、可融合多图并结合世界知识。\", \"en\": \"Google Nano Banana (Gemini image models) for image generation and multi-turn conversational editing, with clear text rendering and strong subject consistency; the Pro version reaches 4K and blends multiple images with world knowledge.\"}","tags":"图像生成,文生图","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":0.06,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"flux-klein-2","description":"{\"zh\": \"Black Forest Labs FLUX.2 [klein](2026-01 发布),FLUX 家族里最快的一档,亚秒级完成生成与编辑。提供 4B 与 9B 两个规模,其中 4B 以 Apache 2.0 开源,9B 采用非商用许可。\", \"en\": \"Black Forest Labs FLUX.2 [klein] (Jan 2026), the fastest tier of the FLUX family, generating and editing images in under a second. It comes in 4B and 9B sizes, with the 4B released under Apache 2.0 and the 9B under a non-commercial licence.\"}","tags":"图像生成,文生图,图像编辑,高速,开源","vendor_id":13,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["image-generation","openai"]},{"model_name":"codex-auto-review","description":"{\"zh\": \"面向代码变更的自动审查模型,基于 OpenAI Codex/GPT-5 编码栈,擅长阅读 diff、发现缺陷与提出修改建议,用于 PR 与代码评审的智能体化审查。\", \"en\": \"An automated code-review model built on the OpenAI Codex / GPT-5 coding stack, good at reading diffs, spotting defects and proposing fixes — for agentic review of PRs and code changes.\"}","tags":"编程,代码审查,智能体","vendor_id":2,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default","gpt专用"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.6-flash","description":"{\"zh\": \"Google Gemini 3.6 Flash(2026-07 发布),Flash 档主打速度与性价比,100 万 token 上下文、原生多模态输入。`-high`/`-medium`/`-low` 后缀对应思考强度档位,`-tiered` 为分层计价版本。\", \"en\": \"Google Gemini 3.6 Flash (Jul 2026). The Flash tier targets speed and value, with a 1M-token context and native multimodal input. The `-high` / `-medium` / `-low` suffixes select thinking effort, and `-tiered` is the tiered-pricing variant.\"}","tags":"多模态,长上下文,高速,均衡","vendor_id":4,"quota_type":0,"model_ratio":0.75,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":0.75,"output":3.75,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.8-flash","description":"{\"zh\": \"Google Gemini 3.8 Flash(2026-09-02 发布,直接 GA),发布时被 Google 称为最聪明的 Flash 模型,面向编程、智能体、多模态推理与专业工作流,并针对长周期编程和自主智能体做了调优。100 万 token 输入、6.4 万输出,支持文本/图像/音频/视频/PDF 输入、函数调用、搜索工具与电脑操作;思考强度分 low/medium/high,默认 medium。`-high`/`-medium`/`-low` 后缀对应思考强度档位。\", \"en\": \"Google Gemini 3.8 Flash (September 2, 2026, launched directly as GA), billed at launch as Google's most intelligent Flash model, for coding, agents, multimodal reasoning and professional workflows, and tuned for long-horizon coding and autonomous agents. 1M-token input and 64K output, accepting text, image, audio, video and PDF, with function calling, search as a tool and computer use; thinking levels low/medium/high, medium by default. The `-high` / `-medium` / `-low` suffixes select the thinking level.\"}","tags":"多模态,长上下文,编程,智能体,高速","vendor_id":4,"quota_type":0,"model_ratio":0.75,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":0.75,"output":3.75,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-embedding-001","description":"{\"zh\": \"Google Gemini 文本嵌入模型,将文本转为高质量多语言语义向量,面向语义检索、聚类与 RAG 检索增强。\", \"en\": \"Google Gemini text-embedding models that turn text into high-quality multilingual semantic vectors for retrieval, clustering and RAG.\"}","tags":"文本嵌入,向量检索","vendor_id":4,"quota_type":0,"model_ratio":0.075,"model_price":0,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","embeddings"]},{"model_name":"o4-mini","description":"{\"zh\": \"OpenAI o4-mini 推理模型,在 o 系列里主打性价比:推理能力接近上一代大模型,但速度与价格接近小模型,适合需要推理又要控成本的高频场景。\", \"en\": \"OpenAI o4-mini is the value option in the o-series: reasoning close to the previous generation's larger models at the speed and price of a small one, which suits high-frequency work that still needs reasoning.\"}","tags":"推理,高速,低成本","vendor_id":2,"quota_type":0,"model_ratio":0.55,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.254545,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"tts-1","description":"{\"zh\": \"OpenAI 的经典语音合成系列。`tts-1` 面向实时场景、延迟最低;`-hd` 档音质更好但更慢;带日期后缀的是固定快照版本。\", \"en\": \"OpenAI's classic text-to-speech series. `tts-1` targets real-time use with the lowest latency, the `-hd` tier trades speed for audio quality, and date-suffixed names are pinned snapshots.\"}","tags":"语音合成","vendor_id":2,"quota_type":0,"model_ratio":7.5,"model_price":0,"owner_by":"","completion_ratio":1,"audio_completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","audio"]},{"model_name":"tts-1-hd","description":"{\"zh\": \"OpenAI 的经典语音合成系列。`tts-1` 面向实时场景、延迟最低;`-hd` 档音质更好但更慢;带日期后缀的是固定快照版本。\", \"en\": \"OpenAI's classic text-to-speech series. `tts-1` targets real-time use with the lowest latency, the `-hd` tier trades speed for audio quality, and date-suffixed names are pinned snapshots.\"}","tags":"语音合成","vendor_id":2,"quota_type":0,"model_ratio":15,"model_price":0,"owner_by":"","completion_ratio":1,"audio_completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","audio"]},{"model_name":"wan2.5-i2v-preview","description":"{\"zh\": \"阿里巴巴通义万相(Wan)视频/图像生成模型,支持文生视频、图生视频与视频编辑。\", \"en\": \"Alibaba Tongyi Wan video/image generation, supporting text-to-video, image-to-video and video editing.\"}","tags":"视频生成","vendor_id":12,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"gemini-3.5-flash-lite","description":"{\"zh\": \"Google Gemini 3.5 Flash-Lite(2026-07-21 GA),3.5 系列里最快的一档,面向低延迟与高吞吐的生产负载,如智能体搜索、文档处理、翻译与大规模数据抽取。原生多模态(文本/图像/语音/视频输入),100 万 token 上下文;思考强度 minimal/low/medium/high 可调,默认 medium。\", \"en\": \"Google Gemini 3.5 Flash-Lite (GA July 21, 2026), the fastest model in the 3.5 series, for low-latency, high-throughput production workloads such as agentic search, document processing, translation and large-scale data extraction. Natively multimodal (text, image, speech and video input) with a 1M-token context; thinking levels minimal/low/medium/high, medium by default.\"}","tags":"高速,低成本,多模态,长上下文","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"official_price":{"input":0.3,"output":2.5,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"claude-opus-4-6","description":"{\"zh\": \"Anthropic Claude Opus 4.6（2026-02 发布），为 Opus 层级引入「自适应思考」与力度参数，100 万 token 上下文，通用推理与长程任务表现强劲。\", \"en\": \"Claude Opus 4.6 (Feb 2026), which brought adaptive thinking and the effort parameter to the Opus tier; 1M-token context with strong general reasoning and long-horizon work.\"}","tags":"推理,编程,长上下文","vendor_id":3,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["default","claude kiro反代 0.2倍率","Claude 专用"],"official_price":{"input":5,"output":25,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"Omni-Flash-Ext","description":"{\"zh\": \"Google Gemini omni 1.1 flash-ext 文生视频模型(上游别名 Omni-Flash-Ext)。走统一的视频生成接口,接受 prompt、时长、分辨率与画面比例,也可传参考图做图生视频。Flash 档主打生成速度。\", \"en\": \"Google's Gemini omni 1.1 flash-ext text-to-video model (exposed upstream as Omni-Flash-Ext). It uses the unified video generation endpoint and takes a prompt, duration, resolution and aspect ratio, and can also accept reference images for image-to-video. The Flash tier is built for generation speed.\"}","tags":"视频生成,文生视频,高速","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["svip","default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"gemini-3.5-flash","description":"{\"zh\": \"Google Gemini 3.5 Flash（2026 发布），主打「以 Flash 的速度与价格提供前沿智能」,原生多模态(文本/图像/音频/视频输入)、100 万 token 上下文,默认动态思考,在部分编码与智能体基准上超越上代 3.1 Pro,面向高速高并发的前沿场景。\", \"en\": \"Google Gemini 3.5 Flash (2026), delivering frontier intelligence at Flash speed and price — natively multimodal (text/image/audio/video), 1M-token context, dynamic thinking on by default, beating the prior 3.1 Pro on some coding/agentic benchmarks.\"}","tags":"推理,多模态,高速,长上下文","vendor_id":4,"quota_type":0,"model_ratio":0.75,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":1.5,"output":9,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.7-flash-high","description":"{\"zh\": \"Google Gemini 3.7 Flash(2026-08 发布),是在 3.6 Flash 推理基础上的算法改进而非新的预训练模型,编程与智能体能力提升明显。100 万 token 输入、6.4 万输出,支持文本/图像/视频/音频/PDF 输入、函数调用、Google 搜索工具与电脑操作。`-high`/`-medium`/`-low` 后缀对应不同的思考强度档位。\", \"en\": \"Google Gemini 3.7 Flash (Aug 2026) is an algorithmic improvement on 3.6 Flash's reasoning rather than a new pretrained model, with a clear jump in coding and agentic ability. 1M-token input and 64K output, accepting text, image, video, audio and PDF, with function calling, Google Search as a tool and computer use. The `-high` / `-medium` / `-low` suffixes select thinking effort.\"}","tags":"多模态,长上下文,编程,智能体,高速","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.7-flash-medium","description":"{\"zh\": \"Google Gemini 3.7 Flash(2026-08 发布),是在 3.6 Flash 推理基础上的算法改进而非新的预训练模型,编程与智能体能力提升明显。100 万 token 输入、6.4 万输出,支持文本/图像/视频/音频/PDF 输入、函数调用、Google 搜索工具与电脑操作。`-high`/`-medium`/`-low` 后缀对应不同的思考强度档位。\", \"en\": \"Google Gemini 3.7 Flash (Aug 2026) is an algorithmic improvement on 3.6 Flash's reasoning rather than a new pretrained model, with a clear jump in coding and agentic ability. 1M-token input and 64K output, accepting text, image, video, audio and PDF, with function calling, Google Search as a tool and computer use. The `-high` / `-medium` / `-low` suffixes select thinking effort.\"}","tags":"多模态,长上下文,编程,智能体,高速","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-oss-120b","description":"{\"zh\": \"OpenAI 的开放权重模型 gpt-oss-120b,1170 亿参数 MoE(每 token 激活约 51 亿),Apache 2.0 许可,可自行部署。推理与工具调用能力接近同期中档闭源模型,是私有化部署的常见选择。\", \"en\": \"OpenAI's open-weight gpt-oss-120b, a 117B-parameter MoE activating about 5.1B per token, released under Apache 2.0 for self-hosting. Its reasoning and tool use approach contemporaneous mid-tier closed models, making it a common choice for private deployments.\"}","tags":"开源,推理,编程,高性价比","vendor_id":2,"quota_type":0,"model_ratio":0.5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"kling-omni-video","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"视频生成","vendor_id":6,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"kling-video","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"视频生成","vendor_id":6,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"minimax-h3","description":"{\"zh\": \"MiniMax Hailuo 03(H3)视频生成模型,每次生成自带原生音轨。三种用法:文生视频;图生视频(给一张图作首帧,也可再给一张尾帧图渲染两图之间的转场,画幅跟随输入图);参考生视频(最多 9 张参考图定主体与风格、3 段参考视频定运动、3 段参考音频,在提示词里按出现顺序引用)。时长 5–15 秒;分辨率 480P/768P/2K/4K,其中 480P/768P 为原生生成,2K/4K 由 768P 上采样;画幅 21:9 至 9:16。按输出秒数计费,单价随分辨率变化。\", \"en\": \"MiniMax Hailuo 03 (H3) video generation, with a native audio track on every clip. Three modes: text-to-video; image-to-video (a still becomes the opening frame, or add a last frame to render a transition between two images, with the aspect ratio following the input); and reference-to-video (up to 9 reference images for subject and style, 3 video clips for motion and 3 audio clips, cited in the prompt in order). 5–15 seconds at 480P/768P/2K/4K — 480P and 768P are native, 2K and 4K are upscaled from a 768P base — in aspect ratios from 21:9 to 9:16. Billed per second of output, priced by resolution.\"}","tags":"视频生成,文生视频,图生视频,多模态","vendor_id":21,"quota_type":1,"model_ratio":0,"model_price":0.117,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai-video"],"video_pricing":{"2k":1.183,"480p":0.455,"4k":1.456,"768p":0.546}},{"model_name":"babbage-002","description":"{\"zh\": \"OpenAI 的传统补全(completion)基座模型,不是对话模型,主要用于微调与遗留脚本。能力远低于现役模型,仅为兼容保留。\", \"en\": \"A legacy OpenAI base completion model - not a chat model - mainly used for fine-tuning and legacy scripts. Far weaker than current models and kept only for compatibility.\"}","tags":"小型,低成本","vendor_id":2,"quota_type":0,"model_ratio":0.2,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4","description":"{\"zh\": \"初代 GPT-4,8K 上下文。历史模型,速度与价格都不占优,保留用于兼容老集成;新项目请用 GPT-4o 及以上。\", \"en\": \"The original GPT-4 with an 8K context. A legacy model that is neither fast nor cheap, kept for compatibility with older integrations; new projects should use GPT-4o or newer.\"}","tags":"推理","vendor_id":2,"quota_type":0,"model_ratio":15,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4o-mini-tts","description":"{\"zh\": \"基于 GPT-4o mini 的语音合成模型,除了音色还能按自然语言指令控制语气与说话风格,适合配音、播报与语音助手。\", \"en\": \"A text-to-speech model built on GPT-4o mini. Beyond voice selection it can be steered with natural-language instructions about tone and delivery, which suits narration, announcements and voice assistants.\"}","tags":"语音合成","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.3,"owner_by":"","completion_ratio":0,"audio_ratio":25,"audio_completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","audio"]},{"model_name":"gpt-5.3-codex","description":"{\"zh\": \"GPT-5.3-Codex(2026-02 发布),OpenAI 面向编程场景的专用版本,强化代码生成、仓库检索、终端命令执行与调试,是 Codex 系列客户端的默认主力。\", \"en\": \"GPT-5.3-Codex (Feb 2026), OpenAI's coding-specialised build, tuned for code generation, repository search, running terminal commands and debugging. It is the default workhorse behind the Codex clients.\"}","tags":"编程,智能体,推理","vendor_id":2,"quota_type":0,"model_ratio":0.875,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":1.75,"output":14,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.1-flash-lite-image","description":"{\"zh\": \"Google Nano Banana(Gemini 图像模型),支持图像生成与多轮对话式编辑,文字渲染清晰、主体一致性强;Pro 版最高 4K、可融合多图并结合世界知识。\", \"en\": \"Google Nano Banana (Gemini image models) for image generation and multi-turn conversational editing, with clear text rendering and strong subject consistency; the Pro version reaches 4K and blends multiple images with world knowledge.\"}","tags":"图像生成,文生图","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":1.176,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"gpt-image-2-official","description":"{\"zh\": \"OpenAI GPT Image 系列图像生成与编辑模型,文字渲染精准、指令跟随强,支持多分辨率与参考图。\", \"en\": \"OpenAI GPT Image generation and editing models with precise text rendering, strong instruction-following, multiple resolutions and reference images.\"}","tags":"图像生成,文生图","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"],"image_tier_ratios":{"quality:high":18,"quality:medium":4.5,"resolution:2k":2,"resolution:4k":3.4}},{"model_name":"gpt-image-2.5-flare","description":"{\"zh\": \"OpenAI GPT Image 2.5 系列(2026-09 发布)。Flare 面向日常快速出图,是官方给多数场景推荐的默认档,响应更快、多轮编辑的指令跟随更稳、参考图中的主体保持更好。支持文生图与图像编辑,quality 分 low/medium/high/xhigh/max、分辨率 1k/2k/4k,按档位计价。异步任务接口调用。\", \"en\": \"OpenAI GPT Image 2.5 (September 2026). Flare targets fast, high-quality everyday generation and is OpenAI's default pick for most applications, with faster responses, steadier multi-turn instruction following and better subject preservation from reference images. Supports text-to-image and editing, with quality tiers low/medium/high/xhigh/max and 1k/2k/4k resolutions, priced by tier. Async task API.\"}","tags":"图像生成,图像编辑,文生图","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.0435,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"],"image_tier_ratios":{"quality:high":9,"quality:max":36,"quality:medium":2.25,"quality:xhigh":16,"resolution:2k":2,"resolution:4k":3.4}},{"model_name":"tts-1-1106","description":"{\"zh\": \"OpenAI 的经典语音合成系列。`tts-1` 面向实时场景、延迟最低;`-hd` 档音质更好但更慢;带日期后缀的是固定快照版本。\", \"en\": \"OpenAI's classic text-to-speech series. `tts-1` targets real-time use with the lowest latency, the `-hd` tier trades speed for audio quality, and date-suffixed names are pinned snapshots.\"}","tags":"语音合成","vendor_id":2,"quota_type":0,"model_ratio":7.5,"model_price":0,"owner_by":"","completion_ratio":1,"audio_completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","audio"]},{"model_name":"gemini-3-pro-image","description":"{\"zh\": \"Google Nano Banana(Gemini 图像模型),支持图像生成与多轮对话式编辑,文字渲染清晰、主体一致性强;Pro 版最高 4K、可融合多图并结合世界知识。\", \"en\": \"Google Nano Banana (Gemini image models) for image generation and multi-turn conversational editing, with clear text rendering and strong subject consistency; the Pro version reaches 4K and blends multiple images with world knowledge.\"}","tags":"图像生成,文生图","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":0.06,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"gemini-3.1-pro","description":"{\"zh\": \"Google Gemini 3.1 Pro,Pro 档是 Gemini 的高能力档,长上下文与多模态理解突出。`-high`/`-low` 后缀对应不同的思考强度:high 更准但更慢更贵,low 反之。\", \"en\": \"Google Gemini 3.1 Pro. The Pro tier is Gemini's high-capability option, strong at long context and multimodal understanding. The `-high` / `-low` suffixes select thinking effort: high is more accurate but slower and dearer, low the reverse.\"}","tags":"推理,多模态,长上下文,旗舰","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-2.0-flash","vendor_id":4,"quota_type":0,"model_ratio":0.075,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.166667,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"qwen3.7-max","description":"{\"zh\": \"阿里通义 Qwen3.7-Max(2026-05 发布),千问闭源旗舰,100 万 token 上下文,纯文本输入输出,支持思考/非思考两种模式、函数调用与显式上下文缓存。定位智能体时代的通用基座,长于高难度逻辑推理与深度计算。\", \"en\": \"Alibaba Qwen3.7-Max (May 2026), the closed-source Qwen flagship with a 1M-token context and text-only input and output, supporting thinking and non-thinking modes, function calling and explicit context caching. It is positioned as a general base model for the agent era and excels at hard logical reasoning.\"}","tags":"旗舰,推理,长上下文,智能体","vendor_id":12,"quota_type":0,"model_ratio":8.75,"model_price":0,"owner_by":"","completion_ratio":3,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"qwen3.6-plus","description":"{\"zh\": \"阿里通义 Qwen3.6-Plus(2026-04 发布),Plus 档主力商用模型,长上下文与多模态理解兼顾,是 3.7-Plus 的上一代。\", \"en\": \"Alibaba Qwen3.6-Plus (Apr 2026), the Plus-tier commercial workhorse balancing long context and multimodal understanding. It is the generation before 3.7-Plus.\"}","tags":"多模态,长上下文,均衡","vendor_id":12,"quota_type":0,"model_ratio":1.75,"model_price":0,"owner_by":"","completion_ratio":6,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"deepseek-v4-flash-vision-exp","description":"{\"zh\": \"DeepSeek 已退役 V4 Flash 视觉实验版,并把这个旧名称转到 V4.1 Flash。新一代 Flash 已原生支持图片理解,不再需要单独的视觉实验版。100 万 token 上下文、最多 38.4 万 token 输出。新接入建议直接用 deepseek-flash。\", \"en\": \"DeepSeek retired the V4 Flash vision experimental model and routes this legacy name to V4.1 Flash, which now understands images natively, so a separate vision build is no longer needed. 1M-token context and up to 384K output. For new integrations, use deepseek-flash directly.\"}","tags":"视觉,多模态,高速,低成本","vendor_id":1,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":12,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":2,"output":8,"currency":"CNY"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.1-pro-high","description":"{\"zh\": \"Google Gemini 3.1 Pro,Pro 档是 Gemini 的高能力档,长上下文与多模态理解突出。`-high`/`-low` 后缀对应不同的思考强度:high 更准但更慢更贵,low 反之。\", \"en\": \"Google Gemini 3.1 Pro. The Pro tier is Gemini's high-capability option, strong at long context and multimodal understanding. The `-high` / `-low` suffixes select thinking effort: high is more accurate but slower and dearer, low the reverse.\"}","tags":"推理,多模态,长上下文,旗舰","vendor_id":4,"quota_type":0,"model_ratio":0.8,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"official_price":{"input":2,"output":12,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-4-preview","vendor_id":4,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.4","description":"{\"zh\": \"OpenAI GPT-5.4 前沿推理模型（2026-03 发布），原生多模态、100 万 token 上下文，具备原生「电脑操作」与工具检索能力，事实性错误较上代显著下降，适合最苛刻的复杂任务。\", \"en\": \"OpenAI GPT-5.4 frontier reasoning model (Mar 2026) — natively multimodal, 1M-token context, with native computer-use and tool-search and markedly fewer factual errors, for the most demanding tasks.\"}","tags":"推理,多模态,长上下文,电脑操作","vendor_id":2,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":2.5,"output":15,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"kling-image","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"图像生成,文生图","vendor_id":6,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"gemini-3.1-flash-image","description":"{\"zh\": \"Google Nano Banana(Gemini 图像模型),支持图像生成与多轮对话式编辑,文字渲染清晰、主体一致性强;Pro 版最高 4K、可融合多图并结合世界知识。\", \"en\": \"Google Nano Banana (Gemini image models) for image generation and multi-turn conversational editing, with clear text rendering and strong subject consistency; the Pro version reaches 4K and blends multiple images with world knowledge.\"}","tags":"图像生成,文生图","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"kimi-k2.7-code","description":"{\"zh\": \"Kimi K2.7 的编程专用版本,针对代码生成、仓库理解与命令行智能体场景做过强化,适合接进编码 CLI 与 IDE 插件。\", \"en\": \"The coding-specialised build of Kimi K2.7, tuned for code generation, repository understanding and command-line agent workflows, and a good fit behind coding CLIs and IDE extensions.\"}","tags":"编程,智能体,开源","vendor_id":11,"quota_type":0,"model_ratio":3.325,"model_price":0,"owner_by":"","completion_ratio":4.2105,"enable_groups":["default","kimi专用"],"official_price":{"input":6.5,"output":27,"currency":"CNY"},"supported_endpoint_types":["openai"]},{"model_name":"imagen-4.0-apimart","description":"{\"zh\": \"Google Imagen 4 文生图家族(Fast/Standard/Ultra),最高 2K 分辨率,写实度、拼写排版与指令遵循出色,内置 SynthID 隐形水印。\", \"en\": \"Google Imagen 4 text-to-image family (Fast/Standard/Ultra), up to 2K resolution, with excellent photorealism, spelling/typography and prompt adherence, plus SynthID invisible watermarking.\"}","tags":"图像生成,文生图","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":0.05,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["image-generation","openai"]},{"model_name":"viduq3","description":"{\"zh\": \"生数科技 Vidu 视频生成模型,支持文生视频、图生视频与多主体参考,生成快速。\", \"en\": \"Shengshu Vidu video generation for text-to-video and image-to-video with multi-subject reference and fast generation.\"}","tags":"视频生成","vendor_id":7,"quota_type":1,"model_ratio":0,"model_price":0.3,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.91,"480p":0.728,"540p":0.364,"720p":0.728}},{"model_name":"gpt-image-2.5-sunburst","description":"{\"zh\": \"OpenAI GPT Image 2.5 系列(2026-09 发布)。Sunburst 面向对编辑精度要求最高的场景,多轮精修时对画面的控制更稳,适合广告主视觉、产品图这类成品级输出。参数与 Flare 相同:支持参考图与图像编辑,quality 分 low/medium/high/xhigh/max、分辨率 1k/2k/4k,按档位计价。异步任务接口调用。\", \"en\": \"OpenAI GPT Image 2.5 (September 2026). Sunburst is built for workflows where editing precision matters most, holding control steady across multi-round refinements - a fit for campaign key visuals and production-ready product imagery. Same parameters as Flare: reference images and editing, quality tiers low/medium/high/xhigh/max and 1k/2k/4k resolutions, priced by tier. Async task API.\"}","tags":"图像生成,图像编辑,文生图","vendor_id":2,"quota_type":1,"model_ratio":0,"model_price":0.0435,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"],"image_tier_ratios":{"quality:high":9,"quality:max":36,"quality:medium":2.25,"quality:xhigh":16,"resolution:2k":2,"resolution:4k":3.4}},{"model_name":"gemini-3.6-flash-low","description":"{\"zh\": \"Google Gemini 3.6 Flash(2026-07 发布),Flash 档主打速度与性价比,100 万 token 上下文、原生多模态输入。`-high`/`-medium`/`-low` 后缀对应思考强度档位,`-tiered` 为分层计价版本。\", \"en\": \"Google Gemini 3.6 Flash (Jul 2026). The Flash tier targets speed and value, with a 1M-token context and native multimodal input. The `-high` / `-medium` / `-low` suffixes select thinking effort, and `-tiered` is the tiered-pricing variant.\"}","tags":"多模态,长上下文,高速,均衡","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.4-nano","description":"{\"zh\": \"GPT-5.4 系列里最小最快的一档(2026-03 上线,仅通过 API 提供)。单价最低、延迟最小,适合分类、抽取、改写、路由判断这类高频轻量任务。\", \"en\": \"The smallest and fastest tier of the GPT-5.4 series (Mar 2026, API-only). Lowest price and lowest latency, suited to high-frequency lightweight work such as classification, extraction, rewriting and routing decisions.\"}","tags":"高速,低成本,小型","vendor_id":2,"quota_type":0,"model_ratio":0.1,"model_price":0,"owner_by":"","completion_ratio":6.25,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":0.2,"output":1.25,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"o1","description":"{\"zh\": \"OpenAI o1 推理系列,是 o 系列的第一代,通过更长的思考链提升数学与科研类问题的准确率。`-mini` 为低成本档,`-preview` 是早期预览版。已被 o3 / o4 系列取代,保留用于兼容。\", \"en\": \"The OpenAI o1 reasoning series, the first of the o-series, improving accuracy on mathematics and research problems through longer chains of thought. `-mini` is the low-cost tier and `-preview` the early preview. Superseded by the o3 / o4 series and kept for compatibility.\"}","tags":"推理,科研","vendor_id":2,"quota_type":0,"model_ratio":7.5,"model_price":0,"owner_by":"","completion_ratio":4,"cache_ratio":0.5,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"seedance-2.0-face","description":"{\"zh\": \"字节跳动 Doubao Seedance 视频生成模型,支持文生视频/图生视频,2.0 起支持约 2K 分辨率、更长片段与原生同步音频、多素材输入。\", \"en\": \"ByteDance Doubao Seedance video generation, supporting text-to-video and image-to-video; 2.0+ adds ~2K resolution, longer clips, native synchronized audio and multi-input.\"}","tags":"视频生成","vendor_id":9,"quota_type":0,"model_ratio":52,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":4.55,"480p":0.9027,"720p":1.9438}},{"model_name":"grok-4.7","description":"{\"zh\": \"xAI Grok 4.7(2026-09 发布),官方定位为面向编程、智能体任务与知识工作的前沿模型,是 Grok 4.6 的后继。50 万 token 上下文。\", \"en\": \"xAI Grok 4.7 (released September 2026), positioned by xAI as its frontier model for coding, agentic tasks and knowledge work, and the successor to Grok 4.6. 500K-token context.\"}","tags":"前沿,推理,编程,智能体,长上下文","vendor_id":10,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":3,"cache_ratio":0.25,"enable_groups":["default"],"official_price":{"input":2,"output":6,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"kling-v3-omni","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"视频生成","vendor_id":6,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.6115,"480p":0.6115,"4k":3.8999,"720p":0.6115}},{"model_name":"wan2.7-videoedit","description":"{\"zh\": \"阿里巴巴通义万相(Wan)视频/图像生成模型,支持文生视频、图生视频与视频编辑。\", \"en\": \"Alibaba Tongyi Wan video/image generation, supporting text-to-video, image-to-video and video editing.\"}","tags":"视频生成","vendor_id":12,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.9974,"480p":0.6042,"720p":0.6042}},{"model_name":"seedream-5-0-lite","description":"{\"zh\": \"字节 Seedream 5.0 的轻量档,画质接近 Pro 档而速度更快、单价更低,适合草稿迭代与大批量出图。\", \"en\": \"The lightweight tier of ByteDance Seedream 5.0, close to the Pro tier in quality but faster and cheaper, suited to draft iteration and high-volume generation.\"}","tags":"图像生成,文生图,高速,低成本","vendor_id":9,"quota_type":1,"model_ratio":0,"model_price":0.2075,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"audio1.0","description":"{\"zh\": \"Vidu 文生音效模型:输入文字描述,生成 2~10 秒的音效,可以精确控制时长和每段声音出现的时间(如「[1s-4s] 屋顶上的雨声」),也能叠加多个音效,适合给视频、游戏与短片配环境音和音效。\", \"en\": \"Vidu's text-to-audio sound-effect model: describe a sound in text and get 2-10 seconds of audio, with precise control over duration and when each sound occurs (e.g. '[1s-4s] raindrops on a rooftop') and support for layering multiple effects, for ambience and sound design in videos, games and shorts.\"}","tags":"创作","vendor_id":7,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","audio"]},{"model_name":"gpt-5-pro","description":"{\"zh\": \"GPT-5 系列的 Pro 档,推理最深的一档,面向高难度数学、科研与复杂工程问题。慢且贵,按需使用。\", \"en\": \"The Pro tier of the GPT-5 series, its deepest-reasoning option, aimed at hard mathematics, research and complex engineering problems. Slow and expensive - use it selectively.\"}","tags":"推理,顶配,科研","vendor_id":2,"quota_type":0,"model_ratio":7.5,"model_price":0,"owner_by":"","completion_ratio":8,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"kling-motion-control","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"视频生成","vendor_id":6,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"o3-pro","description":"{\"zh\": \"OpenAI o3 推理系列,把算力花在回答前的思考上,擅长数学、科研与复杂多步问题。`-mini` 是低成本快速档,`-pro` 思考最久、准确率最高但最慢最贵。\", \"en\": \"The OpenAI o3 reasoning series spends compute thinking before answering, and excels at mathematics, research and complex multi-step problems. `-mini` is the cheap fast tier; `-pro` thinks longest for the highest accuracy but is the slowest and most expensive.\"}","tags":"推理,科研","vendor_id":2,"quota_type":0,"model_ratio":10,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai-response"]},{"model_name":"viduq2","description":"{\"zh\": \"生数科技 Vidu 视频生成模型,支持文生视频、图生视频与多主体参考,生成快速。\", \"en\": \"Shengshu Vidu video generation for text-to-video and image-to-video with multi-subject reference and fast generation.\"}","tags":"视频生成","vendor_id":7,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"claude-sonnet-4-6","description":"{\"zh\": \"Anthropic 上一代中端模型 Claude Sonnet 4.6（2026-02 发布），速度与成本均衡、100 万上下文，适合高并发生产与智能体编程（已被近 Opus 水准的 Sonnet 5 取代）。\", \"en\": \"Anthropic's previous mid-tier model, Claude Sonnet 4.6 (Feb 2026) — balanced speed and cost, 1M context, for high-volume production and agentic coding (superseded by the near-Opus Sonnet 5).\"}","tags":"推理,编程,均衡,高性价比","vendor_id":3,"quota_type":0,"model_ratio":1.5,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"enable_groups":["Claude 专用","default","claude kiro反代 0.2倍率"],"official_price":{"input":3,"output":15,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"kling-v3","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"视频生成","vendor_id":6,"quota_type":1,"model_ratio":0,"model_price":0.4,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.6115,"480p":0.6115,"4k":3.8999,"720p":0.6115}},{"model_name":"qwen-image-2.0-pro","description":"{\"zh\": \"阿里巴巴 Qwen-Image 2.0 统一图像生成+编辑模型,原生 2K 输出,中英双语排版/海报/信息图出色,蒸馏为 4 步快速生成,开源。\", \"en\": \"Alibaba Qwen-Image 2.0, a unified image generation and editing model with native 2K output, excellent Chinese/English typography, posters and infographics, distilled to a 4-step fast path; open-source.\"}","tags":"图像生成,文生图","vendor_id":12,"quota_type":1,"model_ratio":0,"model_price":0.455,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"wan2.7-image-pro","description":"{\"zh\": \"阿里巴巴通义万相(Wan)视频/图像生成模型,支持文生视频、图生视频与视频编辑。\", \"en\": \"Alibaba Tongyi Wan video/image generation, supporting text-to-video, image-to-video and video editing.\"}","tags":"图像生成,文生图","vendor_id":12,"quota_type":1,"model_ratio":0,"model_price":0.495,"owner_by":"","completion_ratio":0,"enable_groups":["svip","default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"grok-image","description":"{\"zh\": \"xAI 的图像生成模型,风格取向偏写实,对时事与流行文化类提示词的理解较好。\", \"en\": \"xAI's image generation model. It leans photorealistic and handles prompts about current events and popular culture comparatively well.\"}","tags":"图像生成,文生图","vendor_id":10,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default"],"supported_endpoint_types":["openai","image-generation"]},{"model_name":"grok-imagine-video-1.5","description":"{\"zh\": \"xAI 视频模型 Grok Imagine Video 1.5(2026-06 GA,Aurora-2 引擎)。最长 15 秒,原生同步音轨与视频同一遍生成,文生视频与图生视频都支持。发布时位列 Image-to-Video Arena 榜首。\", \"en\": \"xAI's video model Grok Imagine Video 1.5 (GA June 2026, Aurora-2). Up to 15 seconds with a natively synchronized audio track generated in the same pass; supports both text-to-video and image-to-video. Ranked #1 on the Image-to-Video Arena at launch.\"}","tags":"视频生成,文生视频,图生视频,多模态","vendor_id":10,"quota_type":1,"model_ratio":0,"model_price":0.12,"owner_by":"","completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"gpt-5.6-sol","description":"{\"zh\": \"OpenAI GPT-5.6 系列的旗舰推理模型（2026-07 发布），为准确度优先、不计速度与成本的最难场景而生——深度编程、复杂智能体任务、安全研究。原生支持文本+视觉输入，约 105 万 token 上下文，新增 max 推理档与协调多个并行子智能体的 ultra 模式。\", \"en\": \"OpenAI's GPT-5.6 flagship reasoning model (Jul 2026), built for the hardest problems where accuracy beats speed and cost — deep coding, complex agentic work, security research. Native text+vision, ~1.05M-token context, with a new 'max' reasoning effort and an 'ultra' mode that coordinates parallel subagents.\"}","tags":"推理,多模态,编程,旗舰","vendor_id":2,"quota_type":0,"model_ratio":2.5,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.12,"enable_groups":["gpt专用","default"],"official_price":{"input":4,"output":20,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-3-flash-preview","description":"{\"zh\": \"Google Gemini 3 Flash（2025 末发布),以 Flash 档的速度与成本提供接近 Pro 级的智能,原生多模态、100 万 token 上下文,思考力度可调,视觉/空间推理与智能体编码显著增强。\", \"en\": \"Google Gemini 3 Flash (late 2025), offering near-Pro intelligence at Flash speed and cost, natively multimodal with 1M-token context, adjustable thinking and much-improved visual/spatial reasoning and agentic coding.\"}","tags":"推理,多模态,高速,长上下文","vendor_id":4,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":0.5,"output":3,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"kimi-k2.6","description":"{\"zh\": \"月之暗面 Kimi K2 系列开源大模型,以 MoE 架构与长上下文见长,擅长编程与智能体任务。`-instruct` 为指令对话版,`-thinking` 为开启深度思考的版本,小数点后的版本号(2.5/2.6/2.7)代表迭代次序,数字越大越新。\", \"en\": \"Moonshot's open-source Kimi K2 series, built on MoE with long context and strong at coding and agentic tasks. `-instruct` is the instruction-tuned chat build and `-thinking` enables deep reasoning; the decimal version numbers (2.5 / 2.6 / 2.7) mark successive iterations, higher being newer.\"}","tags":"开源,编程,智能体,长上下文","vendor_id":11,"quota_type":0,"model_ratio":3.325,"model_price":0,"owner_by":"","completion_ratio":4.2105,"enable_groups":["default","kimi专用"],"official_price":{"input":6.5,"output":27,"currency":"CNY"},"supported_endpoint_types":["openai"]},{"model_name":"kimi-k2-instruct","description":"{\"zh\": \"月之暗面 Kimi K2 系列开源大模型,以 MoE 架构与长上下文见长,擅长编程与智能体任务。`-instruct` 为指令对话版,`-thinking` 为开启深度思考的版本,小数点后的版本号(2.5/2.6/2.7)代表迭代次序,数字越大越新。\", \"en\": \"Moonshot's open-source Kimi K2 series, built on MoE with long context and strong at coding and agentic tasks. `-instruct` is the instruction-tuned chat build and `-thinking` enables deep reasoning; the decimal version numbers (2.5 / 2.6 / 2.7) mark successive iterations, higher being newer.\"}","tags":"开源,编程,智能体,长上下文","vendor_id":11,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["svip","default"],"supported_endpoint_types":["openai"]},{"model_name":"kling-video-o1","description":"{\"zh\": \"快手可灵(Kling)视频/图像生成套件,3.0 支持最高 4K/60fps、多语言同步配音与分镜/运镜控制,涵盖文生视频、图生视频、特效、对口型等能力。\", \"en\": \"Kuaishou Kling video/image generation suite; 3.0 supports up to 4K/60fps, native multi-language synchronized audio and shot/camera control, spanning text-to-video, image-to-video, effects, lip-sync and more.\"}","tags":"视频生成","vendor_id":6,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":0.6115,"480p":0.6115,"720p":0.6115}},{"model_name":"claude-sonnet-5-5","description":"{\"zh\": \"Anthropic Claude Sonnet 5.5(2026-09-28 发布),官方定位为「速度与智能的最佳组合」。100 万 token 上下文、单次最多输出 12.8 万 token,知识截止 2026-06;自适应思考,默认力度 high。价格与 Sonnet 5 相同,但输出速度快 30% 以上、单个任务成本最多低 30%,擅长范围明确的日常任务、修复 bug,以及做出精良的文档、表格与幻灯片。注意:不支持非默认的 temperature / top_p / top_k(本站网关会自动去掉),也不支持强制指定 tool_choice。\", \"en\": \"Anthropic Claude Sonnet 5.5 (released September 28, 2026), positioned as the best combination of speed and intelligence. 1M-token context with up to 128K output per response and a June 2026 knowledge cutoff; adaptive thinking with high default effort. Same price as Sonnet 5, but over 30% faster output and up to 30% lower cost per task; strongest at well-scoped everyday tasks, bug fixing, and polished documents, spreadsheets and slides. Note: non-default temperature / top_p / top_k are not supported (this gateway strips them automatically), and forced tool_choice is not supported.\"}","tags":"推理,编程,智能体,高速,高性价比","vendor_id":3,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":5,"cache_ratio":0.1,"create_cache_ratio":1.25,"enable_groups":["Claude 专用","claude kiro反代 0.2倍率","claude max","default"],"official_price":{"input":2,"output":10,"currency":"USD"},"supported_endpoint_types":["anthropic","openai"]},{"model_name":"gemini-3.6-flash-medium","description":"{\"zh\": \"Google Gemini 3.6 Flash(2026-07 发布),Flash 档主打速度与性价比,100 万 token 上下文、原生多模态输入。`-high`/`-medium`/`-low` 后缀对应思考强度档位,`-tiered` 为分层计价版本。\", \"en\": \"Google Gemini 3.6 Flash (Jul 2026). The Flash tier targets speed and value, with a 1M-token context and native multimodal input. The `-high` / `-medium` / `-low` suffixes select thinking effort, and `-tiered` is the tiered-pricing variant.\"}","tags":"多模态,长上下文,高速,均衡","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.6-flash-tiered","description":"{\"zh\": \"Google Gemini 3.6 Flash(2026-07 发布),Flash 档主打速度与性价比,100 万 token 上下文、原生多模态输入。`-high`/`-medium`/`-low` 后缀对应思考强度档位,`-tiered` 为分层计价版本。\", \"en\": \"Google Gemini 3.6 Flash (Jul 2026). The Flash tier targets speed and value, with a 1M-token context and native multimodal input. The `-high` / `-medium` / `-low` suffixes select thinking effort, and `-tiered` is the tiered-pricing variant.\"}","tags":"多模态,长上下文,高速,均衡","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4-vision-preview","description":"{\"zh\": \"GPT-4 的视觉预览版,是 OpenAI 早期支持图像输入的模型。已被 GPT-4o 全面取代,仅为兼容老集成保留。\", \"en\": \"The vision preview build of GPT-4, OpenAI's early image-input model. It has been superseded by GPT-4o and is kept only for compatibility with older integrations.\"}","tags":"视觉,多模态","vendor_id":2,"quota_type":0,"model_ratio":5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-4o-audio-preview","description":"{\"zh\": \"GPT-4o 的语音预览版,可以直接以音频作为输入和输出,在一次调用里完成听与说,适合语音对话应用。\", \"en\": \"The audio preview build of GPT-4o, able to take audio as both input and output so that listening and speaking happen in a single call - suited to voice conversation applications.\"}","tags":"多模态,语音合成,语音识别","vendor_id":2,"quota_type":0,"model_ratio":1.25,"model_price":0,"owner_by":"","completion_ratio":4,"audio_ratio":16,"enable_groups":["default"],"supported_endpoint_types":["openai","audio"]},{"model_name":"text-embedding-ada-002","description":"{\"zh\": \"OpenAI 文本嵌入模型,把文本编码为向量用于语义检索、相似度、聚类与 RAG;3-large 精度最高,3-small 更具性价比,ada-002 为经典版。\", \"en\": \"OpenAI text-embedding models that encode text into vectors for semantic search, similarity, clustering and RAG; 3-large is highest quality, 3-small is more cost-effective, ada-002 is the classic version.\"}","tags":"文本嵌入,向量检索","vendor_id":2,"quota_type":0,"model_ratio":0.05,"model_price":0,"owner_by":"","completion_ratio":0,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","embeddings"]},{"model_name":"viduq3-mix","description":"{\"zh\": \"生数科技 Vidu 视频生成模型,支持文生视频、图生视频与多主体参考,生成快速。\", \"en\": \"Shengshu Vidu video generation for text-to-video and image-to-video with multi-subject reference and fast generation.\"}","tags":"视频生成","vendor_id":7,"quota_type":0,"model_ratio":0.75,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":1.092,"480p":0.91,"720p":0.91}},{"model_name":"deepseek-v4.1-flash","description":"{\"zh\": \"DeepSeek V4.1 Flash(2026-09-10 发布),与官方调用名 deepseek-flash 为同一模型。100 万 token 上下文、最多 38.4 万 token 输出,原生支持图片理解,支持思考与非思考两种模式、JSON 输出和工具调用。响应快、成本低,适合高并发、对成本敏感的推理与编程场景。\", \"en\": \"DeepSeek V4.1 Flash (released September 10, 2026), the same model as the official API name deepseek-flash. 1M-token context and up to 384K output, with native image understanding, thinking and non-thinking modes, JSON output and tool calls. Fast and low-cost for high-concurrency, cost-sensitive reasoning and coding.\"}","tags":"推理,编程,高速,低成本,视觉","vendor_id":1,"quota_type":0,"model_ratio":0.25,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.08,"enable_groups":["default"],"official_price":{"input":2,"output":8,"currency":"CNY"},"supported_endpoint_types":["openai"]},{"model_name":"gemini-3.7-flash","description":"{\"zh\": \"Google Gemini 3.7 Flash(2026-08 发布),是在 3.6 Flash 推理基础上的算法改进而非新的预训练模型,编程与智能体能力提升明显。100 万 token 输入、6.4 万输出,支持文本/图像/视频/音频/PDF 输入、函数调用、Google 搜索工具与电脑操作。`-high`/`-medium`/`-low` 后缀对应不同的思考强度档位。\", \"en\": \"Google Gemini 3.7 Flash (Aug 2026) is an algorithmic improvement on 3.6 Flash's reasoning rather than a new pretrained model, with a clear jump in coding and agentic ability. 1M-token input and 64K output, accepting text, image, video, audio and PDF, with function calling, Google Search as a tool and computer use. The `-high` / `-medium` / `-low` suffixes select thinking effort.\"}","tags":"多模态,长上下文,编程,智能体,高速","vendor_id":4,"quota_type":0,"model_ratio":0.75,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"official_price":{"input":0.75,"output":3.75,"currency":"USD"},"supported_endpoint_types":["openai"]},{"model_name":"gpt-3.5-turbo-1106","description":"{\"zh\": \"GPT-3.5 Turbo 系列,OpenAI 早期的低成本对话模型。能力已明显落后于现役模型,保留用于兼容老集成与成本极敏感的简单任务。带日期后缀的是固定快照版本。\", \"en\": \"The GPT-3.5 Turbo series, OpenAI's early low-cost chat model. It now lags well behind current models and is kept for compatibility with older integrations and for very cost-sensitive simple tasks. Date-suffixed names are pinned snapshots.\"}","tags":"高速,低成本,小型","vendor_id":2,"quota_type":0,"model_ratio":0.5,"model_price":0,"owner_by":"","completion_ratio":2,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5-mini","description":"{\"zh\": \"GPT-5 系列的中间档,在能力与成本之间取平衡,适合大多数日常任务与中等复杂度的工具调用。\", \"en\": \"The mid tier of the GPT-5 series, balancing capability against cost. It suits most everyday tasks and tool use of moderate complexity.\"}","tags":"均衡,高速,低成本","vendor_id":2,"quota_type":0,"model_ratio":0.125,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.3-chat","description":"{\"zh\": \"GPT-5.3 系列。`-chat` 后缀是面向对话的非推理档,响应快、成本低;`-chat-latest` 跟随 OpenAI 滚动更新到该系列最新的对话版本,不固定在某个快照。\", \"en\": \"The GPT-5.3 series. The `-chat` suffix is the non-reasoning conversational tier: fast and cheap; `-chat-latest` follows OpenAI's rolling updates to the newest chat build in the series rather than pinning to one snapshot.\"}","tags":"推理,编程,均衡","vendor_id":2,"quota_type":0,"model_ratio":1,"model_price":0,"owner_by":"","completion_ratio":6,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"tts-1-hd-1106","description":"{\"zh\": \"OpenAI 的经典语音合成系列。`tts-1` 面向实时场景、延迟最低;`-hd` 档音质更好但更慢;带日期后缀的是固定快照版本。\", \"en\": \"OpenAI's classic text-to-speech series. `tts-1` targets real-time use with the lowest latency, the `-hd` tier trades speed for audio quality, and date-suffixed names are pinned snapshots.\"}","tags":"语音合成","vendor_id":2,"quota_type":0,"model_ratio":15,"model_price":0,"owner_by":"","completion_ratio":1,"audio_completion_ratio":0,"enable_groups":["default"],"supported_endpoint_types":["openai","audio"]},{"model_name":"gemini-3.6-flash-high","description":"{\"zh\": \"Google Gemini 3.6 Flash(2026-07 发布),Flash 档主打速度与性价比,100 万 token 上下文、原生多模态输入。`-high`/`-medium`/`-low` 后缀对应思考强度档位,`-tiered` 为分层计价版本。\", \"en\": \"Google Gemini 3.6 Flash (Jul 2026). The Flash tier targets speed and value, with a 1M-token context and native multimodal input. The `-high` / `-medium` / `-low` suffixes select thinking effort, and `-tiered` is the tiered-pricing variant.\"}","tags":"多模态,长上下文,高速,均衡","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"gpt-5.2-chat-latest","description":"{\"zh\": \"GPT-5.2 系列(2025-12 发布),提供 instant / thinking / Pro 三种模式。`-chat` 后缀为非推理的对话档;`-chat-latest` 跟随滚动更新,不固定快照。\", \"en\": \"The GPT-5.2 series (Dec 2025), offered in instant, thinking and Pro modes. The `-chat` suffix is the non-reasoning conversational tier; `-chat-latest` follows rolling updates rather than pinning to a snapshot.\"}","tags":"推理,均衡","vendor_id":2,"quota_type":0,"model_ratio":0.875,"model_price":0,"owner_by":"","completion_ratio":8,"cache_ratio":0.1,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"minimax-m3","description":"{\"zh\": \"MiniMax M3(2026-06 发布并开源),约 4280 亿总参 / 每 token 激活 230 亿,采用自研 MSA 稀疏注意力架构,最高支持 100 万 token 上下文。原生多模态,支持图片与视频输入,并能操作电脑桌面,主打前沿编程与智能体能力。\", \"en\": \"MiniMax M3 (released and open-sourced June 2026), roughly 428B total parameters activating 23B per token, built on MiniMax's own MSA sparse attention with context up to 1M tokens. It is natively multimodal with image and video input and can drive a computer desktop, targeting frontier coding and agentic ability.\"}","tags":"旗舰,开源,编程,智能体,多模态,长上下文,电脑操作","vendor_id":21,"quota_type":0,"model_ratio":1.05,"model_price":0,"owner_by":"","completion_ratio":4,"enable_groups":["default"],"supported_endpoint_types":["openai"]},{"model_name":"veo3.1-quality","description":"{\"zh\": \"Google Veo 3.1 视频生成模型,支持文生视频/图生视频、高清带音频,含 Fast/Lite/Quality 多档。\", \"en\": \"Google Veo 3.1 video generation for text-to-video and image-to-video, high-definition with audio, in Fast/Lite/Quality tiers.\"}","tags":"视频生成","vendor_id":4,"quota_type":1,"model_ratio":0,"model_price":9.1,"owner_by":"","completion_ratio":0,"enable_groups":["svip","default"],"supported_endpoint_types":["openai","openai-video"]},{"model_name":"veo3.1-quality-official","description":"{\"zh\": \"Google Veo 3.1 视频生成模型,支持文生视频/图生视频、高清带音频,含 Fast/Lite/Quality 多档。\", \"en\": \"Google Veo 3.1 video generation for text-to-video and image-to-video, high-definition with audio, in Fast/Lite/Quality tiers.\"}","tags":"视频生成","vendor_id":4,"quota_type":0,"model_ratio":37.5,"model_price":0,"owner_by":"","completion_ratio":1,"enable_groups":["default","svip"],"supported_endpoint_types":["openai","openai-video"],"video_pricing":{"1080p":1.456,"480p":1.456,"4k":2.912,"720p":1.456},"image_tier_ratios":{"quality:high":18,"quality:medium":4.5,"resolution:2k":2,"resolution:4k":3.4}},{"model_name":"glm-5.2","description":"{\"zh\": \"智谱 GLM-5.2（2026-06 发布），GLM-5 系列开源旗舰，采用稀疏 MoE 架构，100 万 token 上下文、输出可达 13 万 token，提供 High / Max 两档思考力度。面向长程编码智能体与百万级上下文任务，编程与智能体能力突出，开源（MIT）。\", \"en\": \"Zhipu GLM-5.2 (Jun 2026), the open flagship of the GLM-5 series, a sparse MoE with 1M-token context and up to 131K output and High/Max thinking levels. Aimed at long-horizon coding agents and million-token-context work; strong coding and agentic ability; open-source (MIT).\"}","tags":"推理,编程,智能体,长上下文,开源","vendor_id":5,"quota_type":0,"model_ratio":4,"model_price":0,"owner_by":"","completion_ratio":3.5,"cache_ratio":0.25,"enable_groups":["default"],"official_price":{"input":8,"output":28,"currency":"CNY"},"supported_endpoint_types":["openai"]}],"group_ratio":{"auto":1,"claude kiro反代 0.2倍率":0.2,"claude max":1.5,"default":1,"glm专用":0.8,"gpt专用":0.2,"gpt特价":0.12,"grok专用":0.2,"kimi专用":0.5,"vip":1},"pricing_version":"a42d372ccf0b5dd13ecf71203521f9d2","success":true,"supported_endpoint":{"anthropic":{"path":"/v1/messages","method":"POST"},"audio":{"path":"","method":"POST"},"embeddings":{"path":"","method":"POST"},"image-generation":{"path":"","method":"POST"},"openai":{"path":"","method":"POST"},"openai-response":{"path":"/v1/responses","method":"POST"},"openai-video":{"path":"","method":"POST"}},"usable_group":{"auto":"自动优选 · 自动选可用的最低价分组","claude kiro反代 0.2倍率":"Claude Kiro 反代 · 仅 Claude 系列","claude max":"满血claude max","default":"默认 · 全部模型可用","glm专用":"GLM 专用 · 仅 GLM 系列","gpt专用":"GPT 专用 · 仅 GPT 系列","gpt特价":"GPT 特价 · 仅 GPT 系列，最低价但不稳定，追求稳定请用 GPT 专用分组","grok专用":"Grok 专用 · 仅 Grok 系列","kimi专用":"Kimi 专用 · 仅 Kimi 系列","vip":"VIP · 全部模型可用"},"usage_counts":{"claude-fable-5":182,"claude-fable-5-1":441,"claude-haiku-4-5-20251001":304,"claude-opus-4-5-20251101":10,"claude-opus-4-6":1128,"claude-opus-4-7":85,"claude-opus-4-8":561,"claude-opus-5":7683,"claude-opus-5-5":2187,"claude-opus-5.5":578,"claude-sonnet-4-6":696,"claude-sonnet-5":2359,"claude-sonnet-5-5":1288,"codex-auto-review":474,"deepseek-flash":1,"deepseek-v4-flash":14206,"deepseek-v4-flash-vision-exp":165,"deepseek-v4-pro":1861,"deepseek-v4.1-flash":44763,"firefly-gpt-image-2":1,"flux-2-pro":6,"flux-klein-2":26,"flux-kontext-pro":1,"gemini-2.5-flash":8159,"gemini-2.5-flash-image":1,"gemini-2.5-flash-lite":9,"gemini-2.5-pro":5,"gemini-3-flash":60,"gemini-3-flash-preview":401,"gemini-3-pro-image":1,"gemini-3-pro-image-preview":5,"gemini-3-pro-preview":8,"gemini-3.1-flash-image-preview":1,"gemini-3.1-flash-lite":18,"gemini-3.1-flash-lite-image":1,"gemini-3.1-flash-lite-preview":1,"gemini-3.1-pro-high":11,"gemini-3.1-pro-low":3,"gemini-3.1-pro-preview":17,"gemini-3.5-flash":519,"gemini-3.6-flash":4,"gemini-3.7-flash":401,"gemini-3.8-flash":206,"gemini-4-preview":3,"glm-5":15,"glm-5.1":4,"glm-5.2":483,"glm-5.3-flash":415,"gpt-3.5-turbo":11,"gpt-3.5-turbo-1106":1,"gpt-4":2,"gpt-4.1":3,"gpt-4.1-mini":3,"gpt-4.1-nano":7,"gpt-4o":1430,"gpt-4o-mini":6,"gpt-4o-mini-tts":31,"gpt-4o-transcribe":23,"gpt-5":7,"gpt-5-chat-latest":1,"gpt-5-mini":4,"gpt-5-nano":39,"gpt-5-pro":1,"gpt-5.2":6,"gpt-5.3-codex":8,"gpt-5.4":35,"gpt-5.4-mini":13,"gpt-5.4-nano":7,"gpt-5.4-pro":1,"gpt-5.5":16209,"gpt-5.6-luna":216,"gpt-5.6-sol":56846,"gpt-5.6-terra":31480,"gpt-6-astra":89857,"gpt-6-sol":35049,"gpt-6.1-sol":61602,"gpt-image-1.5":3,"gpt-image-2":92,"gpt-image-2-official":25,"gpt-image-2.5-flare":7,"gpt-image-2.5-sunburst":95,"gpt-oss-120b":3,"grok-4.5":4818,"grok-4.6":77,"grok-4.7":1,"grok-imagine-video":3,"grok-imagine-video-1.5":9,"grok-video":16,"imagen-4.0-apimart":2,"kimi-k2.5":37,"kimi-k3":38,"kling-v2-6":6,"kling-v3":3,"minimax-h3":17,"minimax-m3":4,"nano-banana-2":2,"nano-banana-pro":1,"o1-mini":1,"o3":1,"o3-mini":1,"o4-mini":2,"qwen-image-2.0":6,"qwen-image-2.0-pro":1,"qwen3.6-plus":1,"qwen3.7-max":2,"qwen3.7-plus":2,"qwen3.8-flash":47,"seedance-2.0":2,"seedance-2.0-fast":12,"seedream-4-0":1,"seedream-4-5":1,"seedream-5-0-lite":3,"seedream-5-0-pro":1,"text-embedding-3-large":35,"text-embedding-3-small":711,"text-embedding-ada-002":19,"veo3.1-fast":1,"viduq3-turbo":2,"wan2.7-image":1,"wan2.7-image-pro":1,"whisper-1":9,"z-image-turbo":8},"vendors":[{"id":14,"name":"Midjourney","icon":"Midjourney"},{"id":4,"name":"Google","icon":"Gemini.Color"},{"id":5,"name":"Zhipu","icon":"Zhipu.Color"},{"id":6,"name":"Kuaishou","icon":"Kling.Color"},{"id":8,"name":"iFlytek","icon":"Spark.Color"},{"id":9,"name":"ByteDance","icon":"Doubao.Color"},{"id":21,"name":"MiniMax","icon":"Minimax.Color"},{"id":2,"name":"OpenAI","icon":"OpenAI"},{"id":7,"name":"Vidu","icon":"Vidu"},{"id":11,"name":"Moonshot","icon":"Moonshot"},{"id":13,"name":"Black Forest Labs","icon":"Flux.Color"},{"id":15,"name":"Lightricks","icon":"Lightricks"},{"id":18,"name":"智谱","icon":"Zhipu.Color"},{"id":1,"name":"DeepSeek","icon":"DeepSeek.Color"},{"id":3,"name":"Anthropic","icon":"Claude.Color"},{"id":12,"name":"Alibaba","icon":"Qwen.Color"},{"id":16,"name":"SkyReels","icon":"Skywork"},{"id":17,"name":"Baidu","icon":"Wenxin.Color"},{"id":19,"name":"阿里巴巴","icon":"Qwen.Color"},{"id":20,"name":"字节跳动","icon":"Doubao.Color"},{"id":10,"name":"xAI","icon":"XAI"}]}