A broad SDK package can dominate downloads without measuring agent-only use: the Vercel figure is for `ai`.
AI agent frameworks beside LangChain's $3 billion round
LangChain, the company in this trend, makes LangGraph. This list ranks agent frameworks by the sources the AIs used, not by the $3 billion price. LangGraph is first on the download figures. AutoGPT's star count is a different measure.
- No. 1: LangGraph
- 10 places
01Final list
The leading AIs each built their own list of ten independently. Here is one result that is easy to check: every place has a reason, a source, and the list of AIs that included it.
Developer attention: total GitHub stars and six-month star growth
A bright badge means this AI put the entry on its list. A dim badge means it did not.
- 1
LangGraph
The strongest download evidence: 55.1M PyPI downloads in one 30-day window; another source reports 34.5M monthly, so the total depends on the window.
Included it: ChatGPT, Claude, MiMo. Did not include: Meta, Nemotron, Gemini, Nova, StepFun, Kimi, DeepSeek, Wenxin, Qwen.In their lists: ChatGPT: 1 · Claude: 1 · MiMo: 1
What each AI noted 3
- ChatGPTplace 1
About 43.8 million PyPI downloads in one 30-day comparison, despite fewer stars than several older projects.
Source ↗ - Claudeplace 1
About 34.5M monthly downloads against roughly 34k stars is about 1,000 downloads per star (my own division of two figures from different sources). That makes it everyday plumbing, not a showcase.
Source ↗ - MiMoplace 1
33.9k stars but 34.5 million monthly downloads — the biggest bookmark-to-plumbing ratio on the board, and 400 companies ship on its platform.
3 AIs put it here - 2
LangChain
Unconfirmed: Its package recorded 155.9M PyPI downloads in one 30-day window; that broad package count includes automated and dependency pulls.
Included it: ChatGPT, Claude. Did not include: Meta, Nemotron, Gemini, Nova, MiMo, StepFun, Kimi, DeepSeek, Wenxin, Qwen.In their lists: ChatGPT: 2 · Claude: see list
2 AIs put it here - 3
Vercel AI SDK
A supplied tracker reports 118.6M monthly npm downloads, but that is for the broad `ai` package, not agent-only deployments.
Included it: ChatGPT, Gemini, MiMo. Did not include: Meta, Nemotron, Claude, Nova, StepFun, Kimi, DeepSeek, Wenxin, Qwen.In their lists: ChatGPT: 3 · Gemini: 2 · MiMo: 9
What each AI noted 3
- ChatGPTplace 3
A TypeScript package has reached 118.6 million npm downloads in a reported month, making it a major everyday integration layer.
Source ↗ - Geminiplace 2
Averaging 45 million weekly npm pulls by powering the frontend streaming loops of nearly every major web AI interface.
Source ↗ - MiMoplace 9
The web-developer on-ramp: npm-based, browser-native, and the one framework cited alongside Mastra for TypeScript teams.
Source ↗
3 AIs put it here - 4
OpenAI Agents SDK
One supplied comparison reports about 10M monthly installs and 19,000 stars; those figures come from a single secondary source.
Included it: ChatGPT, Gemini, Claude, MiMo. Did not include: Meta, Nemotron, Nova, StepFun, Kimi, DeepSeek, Wenxin, Qwen.In their lists: ChatGPT: 4 · Gemini: 7 · Claude: 2 · MiMo: 2
What each AI noted 4
- ChatGPTplace 4
Its reported 30.6k stars and 4.5 million weekly downloads show substantial reach for a relatively young SDK.
Source ↗ - Geminiplace 7
Rapidly absorbed production mindshare following its launch by formalizing model-native handoffs, structured outputs, and built-in guardrails.
Source ↗ - Claudeplace 2
Released in March 2025, it already has about 10.3M monthly downloads, or 20.25% of the four-framework total.
Source ↗ - MiMoplace 2
19k stars against 10 million monthly installs — roughly 500 installs per star per month, the most-used-per-star framework measured.
Source ↗
4 AIs put it here - 5
CrewAI
About 60,600 stars appear alongside conflicting install counts: 584,900 weekly in one tracker versus 2.4M per 30 days in another.
Included it: ChatGPT, Gemini, Claude, MiMo. Did not include: Meta, Nemotron, Nova, StepFun, Kimi, DeepSeek, Wenxin, Qwen.In their lists: ChatGPT: 5 · Gemini: 3 · Claude: 3 · MiMo: 3
What each AI noted 4
- ChatGPTplace 5
Its roughly 60,600 stars far outstrip its 584,900 weekly downloads in one tracker—a popularity-to-install gap worth treating cautiously.
Source ↗ - Geminiplace 3
Surpassed 60,000 GitHub stars by making multi-agent role-playing intuitive enough for thousands of rapid corporate pilots.
- Claudeplace 3
It has the second-best vote in the brand survey (19.9%), but about 5.2M monthly downloads on roughly 48k to 53k stars is only about 100 downloads per star (my division).
Source ↗ - MiMoplace 3
The multi-agent default in every 2026 comparison — featured in 11-framework, 8-framework and 12-framework roundups alike.
Source ↗
4 AIs put it here - 6
LlamaIndex
Its star counts are substantial, but supplied download estimates range from 413,400 weekly to 12.6M monthly across different sources and windows.
Included it: ChatGPT, Gemini, Claude. Did not include: Meta, Nemotron, Nova, MiMo, StepFun, Kimi, DeepSeek, Wenxin, Qwen.In their lists: ChatGPT: 7 · Gemini: 5 · Claude: see list
What each AI noted 3
- ChatGPTplace 7
Its 476 contributors in one analysis suggest a deeper contributor base than star counts alone reveal.
Source ↗ - Geminiplace 5
Holds 46,000 stars and millions of installs as the primary agentic engine for enterprise retrieval-augmented generation and structured knowledge.
- Claude
Named differently in its own list: open the full list.
see list
3 AIs put it here - 7
Pydantic AI
Its package count varies sharply by package name: one tracker reports 5.3M monthly for `pydantic-ai`, another 23.9M for `pydantic-ai-slim`.
Included it: ChatGPT, Gemini, Claude. Did not include: Meta, Nemotron, Nova, MiMo, StepFun, Kimi, DeepSeek, Wenxin, Qwen.In their lists: ChatGPT: 6 · Gemini: 6 · Claude: 4
What each AI noted 3
- ChatGPTplace 6
The `pydantic-ai-slim` package was reported at 23.9 million monthly downloads, far above the 5.3 million figure for the main package in another tracker.
Source ↗ - Geminiplace 6
Delivering the highest install-to-star conversion for teams prioritizing type safety and strict schema validation in production backend agents.
- Claudeplace 4
About 16k stars, the second-fewest in this list, yet one source reports about 3.8M weekly downloads, which is high for its stars.
Source ↗
3 AIs put it here - 8
AutoGen
A study reports about 182,000 stars, but supplied answers describe maintenance status and give inconsistent or absent current install data.
Included it: ChatGPT, Gemini, Claude, MiMo. Did not include: Meta, Nemotron, Nova, StepFun, Kimi, DeepSeek, Wenxin, Qwen.In their lists: ChatGPT: 9 · Gemini: 4 · Claude: 6 · MiMo: 7
What each AI noted 4
- ChatGPTplace 9
It remains a star heavyweight, but the latest listed release in the November tracker dates to September 30, 2025.
Source ↗ - Geminiplace 4
Pioneered autonomous multi-agent dialogue with 52,000 stars, now bridging enterprise teams into Microsoft's consolidated agent stack.
Source ↗ - Claudeplace 6
It has the highest star count in this group among the Python frameworks (about 55.9k) but only about 404k monthly downloads, which is about 7 per star (my division). It is now in maintenance mode.
Source ↗ - MiMoplace 7
The enterprise incumbent that absorbed AutoGen — chosen wherever .NET and corporate IT already live.
Source ↗
4 AIs put it here - 9
Mastra
A supplied tracker reports 3M weekly downloads, while another reports 8M monthly; npm package and date-window differences prevent a clean comparison.
Included it: ChatGPT, Gemini, MiMo. Did not include: Meta, Nemotron, Claude, Nova, StepFun, Kimi, DeepSeek, Wenxin, Qwen.In their lists: ChatGPT: 8 · Gemini: 10 · MiMo: 6
What each AI noted 3
- ChatGPTplace 8
It has fewer stars than CrewAI but roughly 3 million weekly downloads in one tracker, and its TypeScript focus gives it a different audience.
Source ↗ - Geminiplace 10
TypeScript's rising full-stack agent framework providing deterministic graph execution, state persistence, and automatic eval scoring.
- MiMoplace 6
The TypeScript answer from the Gatsby team — YC W2025 graduate shipping a full-stack agent framework with first-class browser tools.
Source ↗
3 AIs put it here - 10
AutoGPT
A research study reports about 182,000 stars and a one-month gain of 111,967; supplied answers lack a comparable current package-download figure.
Included it: ChatGPT, Gemini. Did not include: Meta, Nemotron, Claude, Nova, MiMo, StepFun, Kimi, DeepSeek, Wenxin, Qwen.In their lists: ChatGPT: 10 · Gemini: 8
What each AI noted 2
- ChatGPTplace 10
A research study found it gained 111,967 stars in one month yet had fewer than nine contributors per 1,000 stars—a vivid warning against equating stars with use.
Source ↗ - Geminiplace 8
Boasts 188,000 GitHub stars as open source's original viral autonomous agent, yet shows the lowest install-to-star ratio in commercial production.
2 AIs put it here
02Worth knowing
What the AIs noticed while building the list.
03Questions people ask
What is the best AI agent framework?
There is no universal winner. The supplied download comparisons put LangGraph high, with reported totals ranging from 34.5M to 55.1M monthly or 30-day downloads. Package windows differ.
Which is the best AI agent framework?
For stateful orchestration, LangGraph is a reasonable choice in this comparison; one supplied tracker reports 55.1M downloads in a 30-day window. That count is not a production-user total.
What are AI agent frameworks?
They are libraries and tools for connecting models to tools, state, workflows, and other agents. The supplied examples include LangGraph, CrewAI, and the OpenAI Agents SDK.
Source ↗Why do framework download figures differ?
Trackers use different date windows and package definitions. For example, supplied answers report Pydantic AI at 5.3M for `pydantic-ai` and 23.9M for `pydantic-ai-slim`.
04How we counted
In plain words, no formulas.
1The question
Top 10 AI Agent Frameworks: GitHub Stars vs Real Installs vs Who Uses Them in Production
2Why this topic
No published table sets attention, installs and production use side by side, so the reader sees for the first time which agent tools are hype and which do the work.
3Where it applies
global, open-source AI agent and orchestration frameworks (LangChain, LangGraph, CrewAI, AutoGPT, LlamaIndex, OpenAI Agents SDK, AutoGen, Mastra, Pydantic AI, Vercel AI SDK and peers)
4What we looked at
- Developer attention: total GitHub stars and six-month star growth
- Actual installs: monthly PyPI and npm downloads, with the share pulled in by other packages noted
- Installs per star, as a ratio that separates showcase projects from everyday plumbing
- Enterprise production use from an independent buyer survey
- A framework must appear in at least two of these measures to be ranked
- GitHub API / AI Agents Info — Repository star snapshots; counts vary by date and repository scope.
- PyPI Stats / AI Agents Info — Package download requests in dated windows; includes automated and dependency pulls, not unique users or production deployments.
- G2 2026 AI Agents Insight Report — Overall B2B agent adoption and production stage; it does not publish comparable framework-level shares.
5Who ranked
The leading AIs, each building its own list of ten independently and naming a source for every place: ChatGPT, Meta, Nemotron, Gemini, Claude, Nova, MiMo, StepFun, Kimi, DeepSeek, Wenxin, Qwen.
6How the places were set
The same entry written differently by different AIs was counted as one. The more AIs put an entry on their list, and the higher they put it, the higher its place. "AI agreement" shows how many of the AIs had it on their own list. Sources are only those the AIs found themselves; nothing was added by hand.
05Each AI's own list
🇺🇸 ChatGPT10 places · 3 notes
The same question, answered by ChatGPT alone. The final list above combines all the AIs.
- 1LangGraph
About 43.8 million PyPI downloads in one 30-day comparison, despite fewer stars than several older projects.
Source ↗ - 2LangChain
Its Python package alone was reported at roughly 157 million downloads over a 30-day window—an enormous footprint, though downloads include automation and dependencies.
Source ↗ - 3Vercel AI SDK
A TypeScript package has reached 118.6 million npm downloads in a reported month, making it a major everyday integration layer.
Source ↗ - 4OpenAI Agents SDK
Its reported 30.6k stars and 4.5 million weekly downloads show substantial reach for a relatively young SDK.
Source ↗ - 5CrewAI
Its roughly 60,600 stars far outstrip its 584,900 weekly downloads in one tracker—a popularity-to-install gap worth treating cautiously.
Source ↗ - 6Pydantic AI
The `pydantic-ai-slim` package was reported at 23.9 million monthly downloads, far above the 5.3 million figure for the main package in another tracker.
Source ↗ - 7LlamaIndex
Its 476 contributors in one analysis suggest a deeper contributor base than star counts alone reveal.
Source ↗ - 8Mastra
It has fewer stars than CrewAI but roughly 3 million weekly downloads in one tracker, and its TypeScript focus gives it a different audience.
Source ↗ - 9AutoGen
It remains a star heavyweight, but the latest listed release in the November tracker dates to September 30, 2025.
Source ↗ - 10AutoGPT
A research study found it gained 111,967 stars in one month yet had fewer than nine contributors per 1,000 stars—a vivid warning against equating stars with use.
Source ↗
What ChatGPT noticed
The largest download totals are not necessarily the most agent-specific: the Vercel figure is for the broad `ai` SDK, while LangChain and LangGraph have their own Python and JavaScript packages.
AutoGPT’s star spike paired with low contributor density is a striking example of attention failing to prove ongoing use or maintenance.
The broad production signal is real but does not identify a winning framework: G2 reported 66% of companies run agents in limited or full production, without a framework-by-framework breakdown in the report announcement.
🇺🇸 Gemini10 places · 1 notes
The same question, answered by Gemini alone. The final list above combines all the AIs.
- 1LangGraph (LangChain)
Converting 47 million monthly PyPI downloads into an undisputed enterprise production backbone with cyclical graph state control.
Source ↗ - 2Vercel AI SDK
Averaging 45 million weekly npm pulls by powering the frontend streaming loops of nearly every major web AI interface.
Source ↗ - 3CrewAI
Surpassed 60,000 GitHub stars by making multi-agent role-playing intuitive enough for thousands of rapid corporate pilots.
- 4AutoGen (Microsoft Agent Framework)
Pioneered autonomous multi-agent dialogue with 52,000 stars, now bridging enterprise teams into Microsoft's consolidated agent stack.
Source ↗ - 5LlamaIndex
Holds 46,000 stars and millions of installs as the primary agentic engine for enterprise retrieval-augmented generation and structured knowledge.
- 6Pydantic AI
Delivering the highest install-to-star conversion for teams prioritizing type safety and strict schema validation in production backend agents.
- 7OpenAI Agents SDK
Rapidly absorbed production mindshare following its launch by formalizing model-native handoffs, structured outputs, and built-in guardrails.
Source ↗ - 8AutoGPT
Boasts 188,000 GitHub stars as open source's original viral autonomous agent, yet shows the lowest install-to-star ratio in commercial production.
- 9Agno (formerly Phidata)
Gained 36,000 stars by focusing on lightweight, memory-first autonomous agents that execute multi-modal tasks with minimal overhead.
- 10Mastra
TypeScript's rising full-stack agent framework providing deterministic graph execution, state persistence, and automatic eval scoring.
What Gemini noticed
Ecosystem Dependency Inflation: Over 60% of LangChain core PyPI installs originate from automated CI test runners or secondary dependencies in third-party libraries rather than direct application code.
🇺🇸 Claude6 places · 3 notes
The same question, answered by Claude alone. The final list above combines all the AIs.
- 1LangGraph / LangChain
About 34.5M monthly downloads against roughly 34k stars is about 1,000 downloads per star (my own division of two figures from different sources). That makes it everyday plumbing, not a showcase.
Source ↗ - 2OpenAI Agents SDK
Released in March 2025, it already has about 10.3M monthly downloads, or 20.25% of the four-framework total.
Source ↗ - 3CrewAI
It has the second-best vote in the brand survey (19.9%), but about 5.2M monthly downloads on roughly 48k to 53k stars is only about 100 downloads per star (my division).
Source ↗ - 4Pydantic AI
About 16k stars, the second-fewest in this list, yet one source reports about 3.8M weekly downloads, which is high for its stars.
Source ↗ - 5Dify
At 153,762 stars it holds 34.86% of all stars in the ten-project snapshot, but the source gives no comparable download figure.
Source ↗ - 6AutoGen
It has the highest star count in this group among the Python frameworks (about 55.9k) but only about 404k monthly downloads, which is about 7 per star (my division). It is now in maintenance mode.
Source ↗
What Claude noticed
LangGraph holds 8.54% of stars but 67.84% of measured downloads, while Dify holds 34.86% of stars and has no comparable download figure. Stars and installs measure different things.
AutoGen was the best-known framework by stars, but it had only 1.68% of downloads and is now in maintenance mode.
The OpenAI Agents SDK reached roughly 10M monthly downloads within about 18 months of release.
🇨🇳 MiMo10 places · 3 notes
The same question, answered by MiMo alone. The final list above combines all the AIs.
- 1LangGraph (2026)
33.9k stars but 34.5 million monthly downloads — the biggest bookmark-to-plumbing ratio on the board, and 400 companies ship on its platform.
- 2OpenAI Agents SDK (2026)
19k stars against 10 million monthly installs — roughly 500 installs per star per month, the most-used-per-star framework measured.
Source ↗ - 3CrewAI (2026)
The multi-agent default in every 2026 comparison — featured in 11-framework, 8-framework and 12-framework roundups alike.
Source ↗ - 4LangChain (2026)
Still the most recognized name — and the only framework whose own vendor page compares it honestly against six rivals.
Source ↗ - 5Dify (2026)
The star leader with 144k GitHub stars — more than four times LangGraph's — while being a platform, not a library.
Source ↗ - 6Mastra (2026)
The TypeScript answer from the Gatsby team — YC W2025 graduate shipping a full-stack agent framework with first-class browser tools.
Source ↗ - 7Microsoft Agent Framework / AutoGen (2026)
The enterprise incumbent that absorbed AutoGen — chosen wherever .NET and corporate IT already live.
Source ↗ - 8Google ADK (2026)
The quiet entrant in every 2026 list — picked not for stars but for being the safe corporate choice next to Gemini.
Source ↗ - 9Vercel AI SDK / eve (2026)
The web-developer on-ramp: npm-based, browser-native, and the one framework cited alongside Mastra for TypeScript teams.
Source ↗ - 10Pydantic AI (2026)
The validation-first framework: type-safety as the selling point, loved by the same teams that already trust Pydantic.
Source ↗
What MiMo noticed
The star board and the install board are different universes: Dify leads stars at 144k, yet LangGraph runs 34.5–71.6M monthly downloads on 33.9k stars — popularity on GitHub is not usage in pip.
Downloads disagree by a factor of two for the same package: Firecrawl reports LangGraph at 34.5M monthly, Kernel measured ~71.6M in its 30-day window — measurement month and counter (PyPI vs pepy) change the headline.
Production is now the norm, not the pilot: G2 finds 57% of companies run agents in production and OutSystems reports 96% of organizations use agents — while Gartner warns 40% of agent projects will still be canceled by end-2027.