社区使用量(谁在被用)
信源:OpenRouter 公开用量榜 (CC BY 4.0)· 批次 2026年10月8日。 这是使用量,不是能力排名 —— 用 token 量给模型排序等于奖励大流量应用(聊天壳、批处理爬虫也算), 所以本页单独成块,绝不参与任何评测分排序;选型请看 场景子榜单与价格。
用量前 10 名 共 545 条 · 另有 40 条折叠 · 只取 standard 通道(限免/批处理通道不并入,否则同一模型重复计 token)
| # | 模型(信源原样) | 厂商 | token 量 | 周环比 | 请求数 | 每次请求 token |
|---|---|---|---|---|---|---|
| 1 | DeepSeek: DeepSeek V4.1 Flash | DeepSeek 深度求索 | 35.59T | +51.5% | 573,611,331 | 62,047 |
| 2 | Z.ai: GLM 5.3 Flash | 智谱 AI | 10.84T | +6.8% | 681,397,279 | 15,910 |
| 3 | Xiaomi: MiMo-V2.6-Flash | Xiaomi | 10.51T | +12.5% | 125,626,112 | 83,632 |
| 4 | Tencent: Hy4 preview | 腾讯 混元 | 7.36T | +8.0% | 65,405,682 | 112,502 |
| 5 | OpenAI: GPT-6 Luna | OpenAI | 6.57T | +15.8% | 306,881,821 | 21,403 |
| 6 | DeepSeek: DeepSeek V4 Flash 0423 | DeepSeek 深度求索 | 5.28T | -22.7% | 401,145,865 | 13,160 |
| 7 | Anthropic: Claude Opus 5.5 | Anthropic | 3.83T | +85.3% | 35,939,421 | 106,563 |
| 8 | typesafe/jev-1.13-20260917 | — | 3.31T | +10.3% | 986,507,318 | 3,358 |
| 9 | Z.ai: GLM 5.3 | 智谱 AI | 3.30T | +29.0% | 66,711,351 | 49,467 |
| 10 | DeepSeek: DeepSeek V4 Flash 0423 | DeepSeek 深度求索 | 3.18T | -0.9% | 375,427,147 | 8,468 |
展开第 11 名起的 40 条(内容一条未删)
| # | 模型(信源原样) | 厂商 | token 量 | 周环比 | 请求数 | 每次请求 token |
|---|---|---|---|---|---|---|
| 11 | Google: Gemini 3.8 Flash | 2.10T | -1.5% | 78,732,577 | 26,679 | |
| 12 | OpenAI: GPT-5.6 Luna | OpenAI | 1.88T | -72.9% | 211,330,947 | 8,917 |
| 13 | MoonshotAI: Kimi K3 | 月之暗面 Kimi | 1.84T | +17.3% | 28,748,005 | 64,126 |
| 14 | Tencent: Hy3 | 腾讯 混元 | 1.82T | -17.1% | 24,774,717 | 73,640 |
| 15 | Z.ai: GLM 5.2 | 智谱 AI | 1.66T | +26.4% | 55,720,385 | 29,820 |
| 16 | OpenAI: GPT-5.6 Sol | OpenAI | 1.64T | +4.0% | 34,218,950 | 47,879 |
| 17 | Meta: Muse Spark 1.3 Contributor | Meta | 1.59T | +6.6% | 40,716,658 | 39,113 |
| 18 | OpenAI: GPT-6.1 Sol | OpenAI | 1.56T | +463.9% | 32,231,531 | 48,422 |
| 19 | Anthropic: Claude Sonnet 5.5 | Anthropic | 1.49T | +268.3% | 26,210,743 | 56,795 |
| 20 | Xiaomi: MiMo-V2.6-Pro | Xiaomi | 1.21T | -2.0% | 26,467,670 | 45,675 |
| 21 | MiniMax: MiniMax M3 | MiniMax | 1.18T | -5.2% | 22,360,125 | 52,933 |
| 22 | DeepSeek: DeepSeek V4 Pro 0423 | DeepSeek 深度求索 | 1.09T | +25.8% | 41,985,058 | 25,901 |
| 23 | OpenAI: GPT-6 Sol | OpenAI | 961.3B | -18.0% | 17,131,744 | 56,110 |
| 24 | OpenAI: GPT-6 Astra | OpenAI | 933.3B | -23.9% | 14,208,211 | 65,684 |
| 25 | Anthropic: Claude Opus 5 | Anthropic | 883.9B | +61.4% | 10,985,122 | 80,461 |
| 26 | Xiaomi: MiMo-V2.5 | Xiaomi | 835.9B | -28.1% | 17,509,660 | 47,740 |
| 27 | Upstage: Solar Mini 4 | Upstage | 828.9B | +586.6% | 14,070,631 | 58,909 |
| 28 | Anthropic: Claude Sonnet 5 | Anthropic | 816.4B | -35.4% | 22,317,245 | 36,583 |
| 29 | Google: Gemini 3 Flash Preview | 811.8B | +9.1% | 77,528,392 | 10,471 | |
| 30 | DeepSeek: DeepSeek V4 Pro 0423 | DeepSeek 深度求索 | 757.7B | +8.8% | 28,895,112 | 26,224 |
| 31 | Upstage: Solar Pro 4 | Upstage | 639.7B | -46.2% | 6,911,065 | 92,556 |
| 32 | StepFun: Step 5 Preview | 阶跃星辰 | 632.2B | — | 3,328,849 | 189,913 |
| 33 | Google: Gemini 2.5 Flash Lite | 596.8B | -17.4% | 443,593,343 | 1,345 | |
| 34 | SpaceXAI: Grok 4.7 | xAI | 580.7B | +41.6% | 8,144,619 | 71,298 |
| 35 | OpenAI: GPT-6 Luna Pro | OpenAI | 577.1B | +38.8% | 11,706,125 | 49,298 |
| 36 | Qwen: Qwen3.8 27B | 阿里云 通义千问 | 534.5B | +45.9% | 24,071,068 | 22,204 |
| 37 | OpenAI: gpt-oss-120b | OpenAI | 493.7B | -16.7% | 125,060,546 | 3,948 |
| 38 | Google: Gemini 3.1 Flash Lite | 481.0B | -6.9% | 114,859,367 | 4,188 | |
| 39 | inclusionAI: Ling 3.1 Flash | 蚂蚁集团 | 479.6B | — | 10,396,410 | 46,135 |
| 40 | Google: Gemini 2.5 Flash | 464.4B | +1.3% | 114,298,384 | 4,063 | |
| 41 | Qwen: Qwen3.8 Flash | 阿里云 通义千问 | 451.1B | -13.0% | 21,610,338 | 20,876 |
| 42 | Google: Gemma 4 31B | 437.6B | -4.1% | 164,487,027 | 2,661 | |
| 43 | Google: Gemini 3.7 Flash | 436.0B | -9.3% | 35,628,919 | 12,237 | |
| 44 | Anthropic: Claude Haiku 5.5 | Anthropic | 380.3B | — | 15,917,746 | 23,894 |
| 45 | Google: Gemma 4 26B A4B | 353.2B | +6.8% | 113,239,232 | 3,119 | |
| 46 | Qwen: Qwen3.7 Flash | 阿里云 通义千问 | 352.9B | -16.9% | 124,232,614 | 2,840 |
| 47 | DeepSeek: DeepSeek V3.2 | DeepSeek 深度求索 | 346.4B | -5.8% | 52,671,868 | 6,576 |
| 48 | qwen/qwen3.8-max-20260902 | — | 325.5B | +34.2% | 5,688,085 | 57,225 |
| 49 | anthropic/claude-4.6-sonnet-20260217 | — | 287.3B | -12.4% | 17,553,470 | 16,365 |
| 50 | Anthropic: Claude Fable 5.1 | Anthropic | 266.2B | -54.0% | 2,474,400 | 107,571 |
第 51 名起的 495 条未在本页展开(全量会撑到几百 KB)—— 完整榜单见信源页面,本站不删数据。
未匹配到本站模型目录的条目按信源标识原样展示(249 条),不隐藏也不猜名字; 周环比为信源口径(其默认统计周期),本站不做二次换算。
谁在调用(应用级消耗)
这是什么:按应用聚合的 token 与请求量 —— 谁在调用这些模型(OpenRouter 近 30 天口径,月榜)。
不代表什么:不是模型能力排名;只覆盖走 OpenRouter 的流量,自建网关与直连 API 不计入,所以它反映的是"这一条路由上的世界",不是全球用量。
排序口径:本表按 token 用量从大到小排列并连续编号
(不采用信源自带的名次字段 —— 那份字段本身有跳号,跳号不代表排名)。
| # | 应用 | tokens | 请求 | 说明 |
|---|---|---|---|---|
| 1 | Hermes Agent | 51.50T | 512,436,626 | Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs pers... |
| 2 | Claude Code | 28.99T | 292,064,823 | Claude Code is Anthropic's agentic coding tool that reads your entire codebase, plans a... |
| 3 | Kilo Code | 21.17T | 260,117,089 | Kilo Code is an open-source AI coding agent that works across VS Code, JetBrains, and C... |
| 4 | Cline | 19.27T | 153,736,073 | Cline is an open-source AI coding agent that lives inside your IDE, autonomously explor... |
| 5 | pi | 11.77T | 109,045,534 | There are many coding agents, but this one is yours. |
| 6 | Codex | 10.89T | 134,613,894 | A coding agent that helps you build and ship with AI |
| 7 | omp | 10.32T | 68,646,960 | |
| 8 | Freebuff | 7.44T | 58,763,066 | Powerful coding models, funded by ads |
| 9 | OpenClaw | 6.01T | 78,716,508 | OpenClaw is an open-source AI agent that connects to your messaging apps and takes real... |
| 10 | DeepSeek Harness | 5.89T | 40,520,917 | |
| 11 | Command Code | 4.19T | 46,844,764 | |
| 12 | OpenHands | 2.92T | 29,993,287 | AI coding agent that can run commands, browse the web, call APIs |
| 13 | ISEKAI ZERO | 1.90T | 36,782,578 | AI adventures. Travel with your favorite characters |
| 14 | ZCode | 1.90T | 10,094,813 | ZCode combines the best AI agents with your existing tools so you can plan, code, revie... |
| 15 | HighLevel | 1.63T | 29,424,194 | |
| 16 | Hello Minds, powered by Ethoswarm | 1.62T | 13,860,453 | |
| 17 | Strix | 1.60T | 13,309,762 | |
| 18 | Lemonade | 1.54T | 31,702,579 | The AI tool for Roblox games. |
| 19 | Nous Research API | 1.28T | 50,533,485 | Nous Research is an AI research team focused on advancing open-source AI. |
| 20 | CodeGPT | 1.18T | 8,829,544 |
开源模型分发量(HuggingFace)
这是什么:开源权重模型的累计下载量与当下趋势分; 只列近 6 个月发布、且属 LLM / 多模态类的模型(不筛的话前几名永远是 2021 年的 embedding 模型,对选型没有参考价值)。 不代表什么:下载 ≠ 生产调用(含 CI、镜像、试玩);只覆盖开源权重,闭源 API 模型不在其中。 批次 2026年10月9日。
累计下载 Top 12
| # | 模型 | 下载 | 类型 |
|---|---|---|---|
| 1 | Qwen/Qwen3.8-27B | 6.84M | image-text-to-text |
| 2 | zai-org/GLM-5.3-Flash | 6.35M | image-text-to-text |
| 3 | RadixArk/Kimi-K3-DSpark | 6.26M | text-generation |
| 4 | audio-cpp/audio.cpp-gguf | 6.23M | text-to-speech |
| 5 | nvidia/Qwen3.6-35B-A3B-NVFP4 | 5.26M | text-generation |
| 6 | ornith-ai/Ornith-1.5-9B-GGUF | 5.15M | text-generation |
| 7 | Qwen/Qwen3.8-27B-FP8 | 4.67M | image-text-to-text |
| 8 | deepseek-ai/DeepSeek-V4-Flash-0731 | 4.52M | text-generation |
| 9 | Qwen/Qwen3.6-35B-A3B-FP8 | 4.46M | image-text-to-text |
| 10 | prism-ml/Ternary-Bonsai-2-27B-gguf | 4.35M | text-generation |
| 11 | lmstudio-community/Qwen3.8-27B-MLX-4bit | 4.17M | image-text-to-text |
| 12 | lmstudio-community/Qwen3.8-27B-MLX-8bit | 3.99M | image-text-to-text |
上升最快 Top 7
| # | 模型 | 趋势分 | 类型 |
|---|---|---|---|
| 1 | autotrust/JEV-27B-VL | 2,591 | image-text-to-text |
| 2 | autotrust/GEV-26B-Decide | 1,682 | text-classification |
| 3 | Cloudflare/clef | 1,280 | image-text-to-text |
| 4 | Aleph-Alpha/Kolibri-1 | 755 | text-generation |
| 5 | abenzerps/Qwen-Image-2.1-Uncensored-GGUF | 673 | text-to-image |
| 6 | jialinyyzz/humanizer | 622 | text-generation |
| 7 | Venastine-Research/Xing4.0-29B-A4B-GGUF | 618 | text-generation |
开发工具 / SDK 采用度(下载量)
这是什么:编码 Agent CLI 与 SDK 的下载量(npm / PyPI,近 30 天)。 不代表什么:这不是调用量 —— 下载包含 CI、镜像重建、容器拉取与试装, 只能粗略代表"有多少项目在接入";同一家公司的多个包会被分开计入,不能相加。 批次 2026年10月9日。
npm(编码 Agent / 协议 SDK)
| # | 包 | 近 30 天下载 |
|---|---|---|
| 1 | @modelcontextprotocol/sdk | 257.67M |
| 2 | @openai/codex | 93.08M |
| 3 | @anthropic-ai/claude-code | 55.55M |
| 4 | opencode-ai | 11.26M |
| 5 | @google/gemini-cli | 1.70M |
| 6 | cline | 330.14K |
PyPI(官方与推理 SDK)
| # | 包 | 近 30 天下载 |
|---|---|---|
| 1 | openai | 295.86M |
| 2 | anthropic | 150.20M |
| 3 | google-genai | 137.61M |
| 4 | litellm | 92.24M |
| 5 | vllm | 1.86M |