Time-series tracking for the AI agent ecosystem
How agent frameworks, MCP servers, and packages grow over time — measured daily. Rankings cross-validate multiple signals so no single gameable metric dominates.
Early development — data is accumulating. Time series grow one day at a time.
Stars are attention (gameable); package downloads are real usage. When a project ranks far apart on the two, that gap is the story a leaderboard hides.
Biggest star gains over the last 7 days — momentum, not cumulative size. This is what a static leaderboard hides.
Ranked by daily tokens actually processed on OpenRouter — the hardest-to-fake usage signal. Chinese models lead.
| # | Name | Tokens/day | Trend |
|---|---|---|---|
| 1 | tencent/hy3-20260706:free | 1.7T | |
| 2 | xiaomi/mimo-v2.5-20260422 | 1.4T | |
| 3 | deepseek/deepseek-v4-flash-20260423 | 805.9B | |
| 4 | z-ai/glm-5.2-20260616 | 658.9B | |
| 5 | minimax/minimax-m3-20260531 | 592.4B | |
| 6 | nvidia/nemotron-3-ultra-550b-a55b-20260604:free | 453.5B | |
| 7 | anthropic/claude-4.7-opus-20260416 | 372.7B | |
| 8 | deepseek/deepseek-v4-pro-20260423 | 370.5B | |
| 9 | anthropic/claude-4.8-opus-20260528 | 361.8B | |
| 10 | openrouter/owl-alpha | 226.8B | |
| 11 | google/gemini-3-flash-preview-20251217 | 155.0B | |
| 12 | anthropic/claude-4.6-sonnet-20260217 | 153.7B | |
| 13 | anthropic/claude-sonnet-5-20260630 | 148.1B | |
| 14 | stepfun/step-3.7-flash-20260528 | 104.9B | |
| 15 | openai/gpt-5.5-20260423 | 99.6B |
Ranked by stars — attention, and easy to game. Commits show whether it is still alive. Real usage is in the download tables below.
Ranked by last-week downloads — actual usage, which is hard to fake.
Ranked by last-month downloads — actual usage.
| # | Name | Downloads | Week | Trend |
|---|---|---|---|---|
| 1 | litellm | 648.8M | 171.5M | |
| 2 | openai | 370.3M | 85.8M | |
| 3 | langchain | 324.5M | 76.7M | |
| 4 | mcp | 297.5M | 68.6M | |
| 5 | google-genai | 288.3M | 65.8M | |
| 6 | browser-use | 174.2M | 10.7M | |
| 7 | langchain-core | 170.2M | 40.2M | |
| 8 | anthropic | 162.2M | 38.8M | |
| 9 | fastmcp | 97.8M | 19.1M | |
| 10 | langchain-openai | 67.9M | 15.8M | |
| 11 | langgraph | 66.7M | 17.1M | |
| 12 | langchain-community | 46.6M | 10.9M | |
| 13 | mistralai | 44.5M | 10.6M | |
| 14 | cohere | 42.3M | 9.2M | |
| 15 | qdrant-client | 24.3M | 5.7M | |
| 16 | langfuse | 23.9M | 5.8M | |
| 17 | langchain-anthropic | 20.9M | 5.2M | |
| 18 | fastapi-mcp | 20.3M | 2.0M | |
| 19 | pydantic-ai | 19.2M | 3.9M | |
| 20 | instructor | 17.3M | 4.3M | |
| 21 | llama-index-core | 15.8M | 3.7M | |
| 22 | chromadb | 13.0M | 3.1M | |
| 23 | crewai | 11.3M | 2.5M | |
| 24 | firecrawl-py | 7.5M | 1.6M | |
| 25 | llama-index | 7.3M | 1.6M | |
| 26 | tavily-python | 6.7M | 1.5M | |
| 27 | ddgs | 6.4M | 1.5M | |
| 28 | dspy | 6.0M | 1.5M | |
| 29 | agno | 2.3M | 481.4k | |
| 30 | e2b-code-interpreter | 2.3M | 617.0k | |
| 31 | crawl4ai | 1.9M | 392.7k | |
| 32 | outlines | 1.9M | 460.3k | |
| 33 | open-webui | 1.5M | 266.6k | |
| 34 | art | 1.3M | 329.6k | |
| 35 | autogen-agentchat | 1.1M | 231.0k | |
| 36 | haystack-ai | 1.0M | 223.0k | |
| 37 | excel-mcp-server | 723.5k | 118.4k | |
| 38 | cloakbrowser | 634.2k | 145.4k | |
| 39 | smolagents | 612.9k | 143.7k | |
| 40 | funasr | 422.2k | 104.5k | |
| 41 | code-review-graph | 342.9k | 91.4k | |
| 42 | agentops | 299.5k | 69.1k | |
| 43 | evo | 223.3k | 56.6k | |
| 44 | cognee | 144.0k | 33.4k | |
| 45 | opensandbox | 109.9k | 32.9k | |
| 46 | jcodemunch-mcp | 88.4k | 21.9k | |
| 47 | mirascope | 85.0k | 16.5k | |
| 48 | financetoolkit | 59.4k | 16.3k | |
| 49 | pageindex | 56.5k | 13.3k | |
| 50 | pocketflow | 40.5k | 8.3k | |
| 51 | langroid | 31.9k | 11.3k | |
| 52 | vision-agents | 31.6k | 9.3k | |
| 53 | metagpt | 25.2k | 3.9k | |
| 54 | guidance | 24.7k | 5.7k |
Ranked by Marketplace installs — real adoption of coding agents, not stars.
| # | Name | Installs | Trend |
|---|---|---|---|
| 1 | GitHub Copilot Chat | 76.6M | |
| 2 | GitHub Copilot | 74.1M | |
| 3 | Claude Code for VS Code | 20.7M | |
| 4 | (Legacy) Tabnine | 9.6M | |
| 5 | Gemini Code Assist | 4.8M | |
| 6 | Cline | 4.7M | |
| 7 | Windsurf Plugin (formerly Codeium): AI Coding Autocomplete and Chat for Python, JavaScript, TypeScript, and more | 3.9M | |
| 8 | Continue - open-source AI code agent | 3.7M | |
| 9 | Qoder CN （Formerly Lingma） - Alibaba Cloud AI Agentic Coding Platform | 2.6M | |
| 10 | Roo Code | 1.8M | |
| 11 | Amazon Q | 1.8M | |
| 12 | Kilo Code: AI Coding Agent, Copilot, and Autocomplete | 1.3M | |
| 13 | CodeGeeX: AI Coding Assistant | 1.3M | |
| 14 | Cody: AI Code Assistant | 865.5k | |
| 15 | Augment: Coding Agent Built for Large, Complex Codebases | 768.7k |
Ranked by ModelScope downloads — real distribution of domestic models (Qwen / DeepSeek / GLM …).
| # | Name | Downloads | Stars | Trend |
|---|---|---|---|---|
| 1 | Qwen2.5-7B-Instruct | 7.1M | 489 | |
| 2 | Qwen3-8B | 7.0M | 307 | |
| 3 | Qwen2.5-72B-Instruct | 4.6M | 219 | |
| 4 | DeepSeek-R1 | 926.3k | 1.4k | |
| 5 | Qwen3-235B-A22B | 890.4k | 229 | |
| 6 | MiniCPM-V-2_6 | 753.8k | 118 | |
| 7 | Meta-Llama-3.1-8B-Instruct | 470.1k | 131 | |
| 8 | glm-4-9b-chat | 412.2k | 248 | |
| 9 | DeepSeek-V3 | 333.5k | 254 | |
| 10 | Baichuan2-13B-Chat | 263.4k | 251 | |
| 11 | GLM-4.6 | 111.0k | 286 | |
| 12 | Kimi-K2-Instruct | 102.6k | 179 | |
| 13 | internlm2_5-7b-chat | 82.1k | 34 | |
| 14 | DeepSeek-V3.1 | 61.6k | 253 | |
| 15 | GLM-4.5 | 30.1k | 402 | |
| 16 | Yi-1.5-9B-Chat | 12.1k | 22 |