Vital signs
Why it's on the table
On the table, Agent Framework (Ag) is seat 9 of 58, in the Agents & Automation family. It is an emerging element — the job is real and here to stay, but the leaderboard still changes quarterly. Choose for this quarter, hold loosely, and watch the changelog. It is optional: plenty of companies run without it — until a specific trigger (scale, regulation, cost, or customers) makes it essential for them. The price of entry is zero, which makes trying it a decision that needs no meeting.
Agent Framework: the top 5 — v2026.Q3
Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.
1LangGraphLangChain
Framework free (MIT) · LangSmith/Platform: Dev $0 + usage · Plus $39/seat/mo · Enterprise customBest for Long-running, stateful agents that must survive restarts, pause for human approval, and resume — the production-grade default.
The category's center of gravity: 71.6M monthly PyPI downloads (Aug 2026), 38.1k stars, and 1.0 stable since Oct 22, 2025 with API stability promised until 2.0 — rare discipline in a churn-heavy category. Checkpointed persistence, human-in-the-loop interrupts, and time-travel debugging are native, not bolted on; the same graph runtime powers deployments at Uber, LinkedIn, Klarna, JPMorgan and Rippling.
Watch The steepest learning curve of the top five — graph ceremony is overkill for simple tool loops, and the ecosystem's gravity pulls you toward paid LangSmith/Platform ($1.50/LCU adds up on chatty agents). Aggregate download counts include heavy CI/mirror traffic — treat absolutes with care.
2OpenAI Agents SDKOpenAI
Free (MIT) · you pay standard OpenAI API token + tool pricing · works with 100+ providers via LiteLLMBest for The shortest path from idea to working agent — minimal primitives (agents, handoffs, guardrails, sessions) with OpenAI's weight behind them.
The canon's 'default path' earned it: 35.5M monthly PyPI downloads and 26.3k stars by Aug 2026, and the April 15, 2026 update added what it lacked — a model-native harness with sandboxed execution (E2B, Modal, Cloudflare, Vercel), workspace manifests, and durable long-horizon runs with snapshotting. Oscar Health and LexisNexis vouched for production viability at launch.
Watch Still pre-1.0 (v0.19.4) after 99 releases — API drift is real and testers hit breaking changes. Provider-agnostic on paper, but tracing, evals, and the new harness all pull you deeper into the OpenAI platform.
3CrewAICrewAI Inc.
OSS free (MIT) · Platform: free tier 50 executions/mo · Enterprise custom (SSO, RBAC, forward-deployed engineers)Best for Role-based multi-agent teams — describe agents as colleagues with roles, goals and tools, and get a working crew in an afternoon.
The most-starred dedicated agent framework (56.4k, Aug 2026) and the one that made multi-agent legible to non-researchers. 2B agent executions in the 12 months to Apr 2026, ~14.3M monthly PyPI downloads, and an enterprise platform (Studio, Control Plane) the company says reaches 65% of the Fortune 500 — a vendor claim, but DocuSign, PepsiCo and Experian are named customers.
Watch The tested weaknesses are consistent: an 856MB install dragging unneeded dependencies, 4.0s median latencies (slowest of the majors bar Claude SDK), and multi-agent chatter that inflates token costs 3–5x. Only $18M raised — thin capitalization next to LangChain's $260M if the category consolidates.
4Pydantic AIPydantic Services
Free (MIT) · optional Logfire observability (free tier, then usage-based)Best for Production Python teams who want agents treated as typed, testable software — validation, dependency injection, structured streaming.
The 'agents are software' thesis, from the team whose validation library sits under every other framework on this page. v1 stable Sept 2025, now v2.24 with durable execution (Temporal/DBOS) and graph support; 16.1M monthly PyPI downloads by Aug 2026 — past CrewAI on installs with a third of the stars. Independent testing found reference agents in ~160 lines vs 280 (LangGraph) and 420 (CrewAI), with whole bug classes caught at type-check time.
Watch Smallest community of the top five (17.5k stars), no visual tooling or hosted runtime — you bring your own deployment story. Moved fast from v1 to v2; 'stable' has meant less here than at LangChain.
16.1M PyPI downloads in the last 30 days; v2.24.0 (Aug 2026) [src] · 17.5k GitHub stars, MIT (Aug 2026); model-agnostic across OpenAI, Anthropic, Gemini, Mistral et al. [src] · v1 stable Sept 2025; reference build ~160 LoC vs 280 (LangGraph) / 420 (CrewAI); adopters incl. Sophos, MindsDB (Apr 2026, single-source) [src]5MastraMastra (YC W25, ex-Gatsby team)
Framework free (Apache-2.0, EE dirs licensed) · Cloud: Starter $0 · Teams $250/mo · Enterprise customBest for TypeScript-first teams building agent products — agents, workflows, memory, evals and a visual studio in one modern TS stack.
The TypeScript answer to LangGraph, and the surprise winner of AgentMail's 9-framework 2026 bake-off (fastest median latency at 2.2s, competitive token overhead). 26.9k stars by Aug 2026, running in production at Replit, SoftBank, Sanity and WorkOS, with workflow suspend/resume, semantic-recall memory and Mastra Cloud shipping monthly (managed sandboxes and rubric scorers landed Aug 2026).
Watch Youngest of the five — the open-core split (Apache core, 'Mastra Enterprise License' ee/ directories) appeared in 2026 and is worth reading before you depend on a feature. TypeScript only; Python shops should look elsewhere.
Agent Framework: the top 8 compared
Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.
| Tool | Language | License | Installs (30d, Aug 2026) | Multi-agent model | Durable state | Models | MCP / A2A | Managed platform |
|---|---|---|---|---|---|---|---|---|
| LangGraph | Python · JS | MIT | 71.6M PyPI | Graphs, supervisor, swarm | Native checkpoints + resume | Any provider | MCP yes · A2A via platform | LangGraph Platform + LangSmith |
| OpenAI Agents SDK | Python · TS | MIT | 35.5M PyPI | Handoffs + guardrails | Sessions; harness snapshots (Apr 2026) | OpenAI-first, 100+ via LiteLLM | MCP yes | OpenAI platform / AgentKit |
| Pydantic AI | Python | MIT | 16.1M PyPI | Handoffs, graphs | Durable exec (Temporal, DBOS) | Any provider | MCP yes | Logfire (observability only) |
| CrewAI | Python | MIT | 14.3M PyPI | Role-based crews + flows | Flow persistence + resume | Any via LiteLLM | MCP yes | CrewAI Platform (AMP) |
| Mastra | TypeScript | Apache-2.0 + EE | ~300k+/wk npm | Agent networks + workflows | Workflow suspend/resume | Any via AI SDK | MCP yes | Mastra Cloud |
| MS Agent Framework | Python · .NET | MIT | young (GA Apr 2026) | Graph, group chat, Magentic-One | Checkpoints + HITL | Azure, OpenAI, Anthropic, Bedrock, Gemini, Ollama | MCP + A2A native | Azure AI Foundry |
| Claude Agent SDK | Python · TS | SDK MIT, service commercial | n/a | Subagents | Sessions, resume/fork | Claude only | MCP native | Anthropic Managed Agents |
| smolagents | Python | Apache-2.0 | n/a | Code-writing agents | Minimal | Any via HF / LiteLLM | MCP yes | Hugging Face ecosystem |
How to choose your agent framework
- If your agent must run for hours, survive a restart, and pause for a human sign-off
- LangGraph — checkpointed persistence and interrupts are the core product, not a plugin, and the 1.0 API-stability promise de-risks the bet.
- If you're OpenAI-committed and want the fewest abstractions between you and a working agent
- OpenAI Agents SDK — ~16 lines to a tool-calling agent, plus the Apr 2026 sandboxed harness for long-horizon work. Accept pre-1.0 API drift as the tax.
- If you need a multi-agent demo for stakeholders this week, or business-process automation with named roles
- CrewAI — fastest concept-to-crew of the majors. Budget for the 3–5x token multiplier that agent chatter brings before you ship it.
- If your company writes TypeScript
- Mastra for agent-first products; Vercel AI SDK's ToolLoopAgent if you're just adding agent features to an existing Next.js/React app.
- If correctness, testing, and maintainability outrank orchestration ceremony
- Pydantic AI — typed outputs, dependency injection, and ~40% less code than LangGraph on reference builds. Also the honest default if a plain 50-line tool loop would do: learn loops, tools and state; the brand names rotate.
Agent Framework: the whole field
21 more tools tracked in this category, including 6 dead, renamed, or sunsetting — a reference that hides the graveyard isn't one. Verified 2026-08-06.
| Tool | Maker | What it is | Entry | Status |
|---|---|---|---|---|
| Microsoft Agent Framework | Microsoft | SK + AutoGen unified; 1.0 GA Apr 3, 2026 with .NET/Python parity, native MCP + A2A, Azure AI Foundry hookup — the .NET-shop default and the near-miss for the top 5 | free (MIT) | active |
| Claude Agent SDK | Anthropic | Claude Code's agent loop as a Python/TS library — tools, subagents, hooks, skills, MCP. Highest capability ceiling, heaviest context footprint (~35k tokens in tests), Claude-only | free SDK + API usage | active |
| Vercel AI SDK | Vercel | The TS toolkit (24.9k stars) with ToolLoopAgent and agent-UI streaming — lightest install of any tested framework; best for adding agents to existing web apps | free (Apache-2.0) | active |
| smolagents | Hugging Face | Agents that write code instead of JSON tool calls — 27.3k stars, barebones by design; great for learning and single-agent automation | free (Apache-2.0) | active |
| Google ADK | Code-first Python/Java toolkit (19.6k stars) behind Vertex AI Agent Builder; hierarchical agents, strong evals — friction outside GCP | free (Apache-2.0) | active | |
| Strands Agents | AWS | Model-driven Python SDK (5.9k stars, v1.40 May 2026); AWS's open framework, pairs with Bedrock AgentCore for hosting | free (Apache-2.0) | active |
| Agno | Agno (ex-Phidata) | 41.6k stars — SDK + AgentOS runtime + UI in one; the strongest 'agent platform in a box' pitch of the independents | free (Apache-2.0) | active |
| Letta | Letta (ex-MemGPT) | Stateful agents with self-editing memory (24k stars) — overlaps the memory element; pick it when persistence IS the product | free + cloud tiers | active |
| LlamaIndex Workflows | LlamaIndex | Event-driven agent workflows atop the RAG library; solid, but the company's center of gravity is LlamaCloud document AI → see Vd | free (MIT) | active |
| Haystack | deepset | Pipeline framework (25.5k stars) with agent support; leans RAG/search more than autonomy | free (Apache-2.0) | active |
| AG2 | community (ex-AutoGen founders) | The fork that kept AutoGen's conversable-agent API alive — ~100k monthly installs, the refuge for teams refusing the MAF migration | free (Apache-2.0) | active |
| AutoGen | Microsoft | Maintenance mode since Apr 2026 — the first major framework sunset; 54k+ developers pointed at Agent Framework, AG2, or a frozen 0.2 | — | sunsetting |
| Semantic Kernel | Microsoft | The enterprise .NET pioneer (28k stars), folded into Agent Framework; v1.x maintained at least a year past MAF GA, but the road ends there | — | sunsetting |
| DSPy | Stanford NLP / community | Programs-not-prompts optimizer, often mislabeled an agent framework — it compiles prompts, it doesn't run tool loops; pair it with one of the above | free (MIT) | active |
| AutoGPT | Significant Gravitas | The 2023 spark (186k stars) pivoted into a low-code agent platform (Polyform Shield license) — belongs closer to Wf now than to this element | free + platform | active |
| BabyAGI | Yohei Nakajima | The 105-line task-loop that co-launched the 2023 agent craze; reframed as an experimental 'self-building' toy in 2024, quiet since | — | fading |
| MetaGPT | FoundationAgents (DeepWisdom) | 'Software company as multi-agent' research hit; team's energy moved to the MGX app-builder product → see Vb | free | fading |
| SuperAGI | SuperAGI | Dev-first framework of 2023 that pivoted to packaged sales/CRM agents; framework repo momentum long gone | — | fading |
| ElizaOS | Eliza Labs | TypeScript agent OS that owns the web3/social-agent niche; big stars, narrow lane | free (MIT) | active |
| CAMEL / OWL | CAMEL-AI.org | Research-grade multi-agent society framework; strong papers, thinner production story | free (Apache-2.0) | active |
| Griptape | Griptape | Python framework + managed cloud with an enterprise-security pitch; development cadence visibly slowed through 2026 | free + cloud | fading |
Agent Framework: the category in numbers
Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.
- AI agents market: $7.84B (2025) → $52.62B (2030) projected, 46.3% CAGR; multi-agent systems segment growing 48.5% [src]
- LangChain Inc. raised $125M at $1.25B (Oct 2025, IVP) — $260M total; the only framework vendor at unicorn scale [src]
- The great consolidation: Microsoft merged Semantic Kernel + AutoGen into Agent Framework 1.0 (GA Apr 3, 2026), putting AutoGen — 2023's hottest framework — into maintenance mode [src]
- Install volume exploded: langgraph 71.6M + openai-agents 35.5M + pydantic-ai 16.1M + crewai 14.3M PyPI downloads in the 30 days to Aug 2026 [src]
- Protocols became table stakes: all 10 leading frameworks support MCP; A2A crossed 150+ adopting orgs by Apr 2026 with native support in Azure AI Foundry, Bedrock AgentCore, and Google Cloud [src]
- Frameworks remain unstable ground: 5 of 9 frameworks threw breaking API changes during one reviewer's 90-run test window (2026) [src]
Agent Framework: method & sources
Ranking criteria: verified install base and star momentum (GitHub/pypistats fetched Aug 6, 2026), API stability track record, production evidence (named customers), and independent hands-on testing (AgentMail 9-framework bake-off, Alice Labs) — not vendor claims alone. Download caveat: pypistats totals include CI and mirror traffic, so absolutes are inflated for every framework; we use them comparatively. This is why we kept pypistats over AgentMarketCap's much lower 'CrewAI 1.3M installs/mo' figure — same signal, different counting; relative order is unchanged. Star-count conflicts among 2026 aggregators (Pydantic AI '16k', LangGraph '9.5k' in one piece) were resolved by fetching GitHub directly. CrewAI's '65% of Fortune 500' is a vendor claim with third-party revenue estimates of only $2.4–3.2M (2025) — believe the adoption, discount the depth. Alice Labs' 'Hermes Agent, 220k stars since Feb 2026' claim is single-source and unverifiable; excluded. Scope pointers: visual/low-code agent builders (Dify, Flowise, Langflow, n8n, OpenAI's AgentKit Agent Builder canvas, AutoGPT Platform) → Wf · Workflow Automation; durable-execution and scheduling infra (Temporal, Inngest) → Or · Orchestration; browser-driving agents → Bu; coding-agent harnesses as products (Claude Code, Codex) → Ca; agent memory layers (Letta's core value, Mem0, Zep) → the memory element. Statuses for long-tail field entries (Griptape, CAMEL, SuperAGI, MetaGPT, ElizaOS, BabyAGI) reflect lighter verification than the majors. Ranking criteria: verified commercial traction, independent satisfaction surveys, agent benchmarks, and founder-fit (price floor, lock-in, surfaces). Editorial, never paid — the charter. Machine-readable twin: ag.json.
All sources (28)
- https://pypistats.org/packages/langgraph
- https://pypistats.org/packages/openai-agents
- https://pypistats.org/packages/pydantic-ai
- https://pypistats.org/packages/crewai
- https://github.com/langchain-ai/langgraph
- https://github.com/crewAIInc/crewAI
- https://github.com/openai/openai-agents-python
- https://github.com/pydantic/pydantic-ai
- https://github.com/mastra-ai/mastra
- https://github.com/huggingface/smolagents
- https://github.com/microsoft/agent-framework
- https://github.com/google/adk-python
- https://github.com/anthropics/claude-agent-sdk-python
- https://www.langchain.com/blog/langchain-langgraph-1dot0
- https://www.langchain.com/pricing
- https://www.crewai.com/pricing
- https://mastra.ai/pricing
- https://sacra.com/c/langchain/
- https://techcrunch.com/2026/04/15/openai-updates-its-agents-sdk-to-help-enterprises-build-safer-more-capable-agents/
- https://openai.com/index/the-next-evolution-of-the-agents-sdk/
- https://visualstudiomagazine.com/articles/2026/04/06/microsoft-ships-production-ready-agent-framework-1-0-for-net-and-python.aspx
- https://agentmarketcap.ai/blog/2026/04/13/microsoft-autogen-maintenance-mode-agent-framework-sunset-2026
- https://www.agentmail.to/blog/best-ai-agent-frameworks-2026
- https://alicelabs.ai/en/insights/best-ai-agent-frameworks-2026
- https://www.marketsandmarkets.com/Market-Reports/ai-agents-market-15761548.html
- https://code.claude.com/docs/en/sdk/overview
- https://www.getpanto.ai/blog/crewai-platform-statistics
- https://www.firecrawl.dev/blog/best-open-source-agent-frameworks
Our take
Frameworks churn fast; the concepts don't. Learn loops, tools, and state once — the brand names will rotate under you.
Combines with
Appears in compounds
Guides
This is element 9 of 58. The table is versioned quarterly — when a tool loses its seat, the changelog records the succession.
Explore the full table →