Every Coding Agent · Tracked
The live map of the agent economy · every brain, every harness, every tool. Capability scores across SWE-bench, GAIA, OSWorld and τ-bench, plus GitHub momentum for the fastest risers.
Hottest this week
· by 7-day star ΔTop 5 fastest-rising open-source agents · refreshed daily
Agent Momentum Quadrant
· capability × velocityX capability score · Y momentum · bubble size stars · color brain family
Best for the job
Top 5 agents for common tasks · rankings refresh every time the data does, no editorial picks.
Capability leaderboard
· 21 agentsTab between categories · sort by any column
| # | Agent | Capability ↓ | Stars | ||
|---|---|---|---|---|---|
| 1 | Claude Code Anthropic | n/a | 112.5k | ||
| 2 | Codex CLI OpenAI | n/a | 74.6k | ||
| 3 | Cursor Anysphere | n/a | n/a | ||
| 4 | Windsurf Codeium | n/a | n/a | ||
| 5 | Devin Cognition | n/a | n/a | ||
| 6 | Manus Butterfly Effect | n/a | n/a | ||
| 7 | Cline Cline | n/a | 60.2k | ||
| 8 | Roo Code Roo Code | n/a | 23.1k | ||
| 9 | Continue Continue | n/a | 32.5k | ||
| 10 | GitHub Copilot GitHub | n/a | n/a | ||
| 11 | GPT Engineer Anton Osika | n/a | 55.2k | ||
| 12 | AutoGPT Significant Gravitas | n/a | 183.3k | ||
| 13 | MetaGPT DeepWisdom | n/a | 66.9k | ||
| 14 | Plandex Plandex | n/a | 15.2k | ||
| 15 | Zed Zed Industries | n/a | 78.9k | ||
| 16 | Tabby TabbyML | n/a | 33.4k | ||
| 17 | Aider Paul Gauthier | 26.3 | 43.2k | ||
| 18 | Jules Google | 52.2 | n/a | ||
| 19 | Sweep Sweep AI | 53.4 | 7.7k | ||
| 20 | SWE-agent Princeton NLP | 66.6 | 19.0k | ||
| 21 | OpenHands All Hands AI | 71.8 | 71.0k |
The Stack
· brain + harness + tools + environmentEvery agent is a recipe · same brain + different harness often means massive score delta
By category
Classified by behavior · autonomous, assistive, vertical, infrastructure
Autonomous
· 9 agentsAgents that run end-to-end without step-by-step human oversight. Plan, act, verify, and ship.
Open-source autonomous software engineer
Agent-computer interfaces for software engineering
GitHub issue to pull request
Asynchronous coding agent from Google Labs
Autonomous software engineer
General-purpose autonomous agent
Specify what you want it to build
The original autonomous GPT agent
Multi-agent collaborative framework
Assistive
· 12 agentsPair-programmer style agents. Human stays in the loop, agent drives the keyboard.
AI pair programming in your terminal
Terminal-first coding agent by Anthropic
OpenAI's terminal coding agent
AI-first code editor
Agentic IDE by Codeium
Autonomous coding agent for VS Code
Autonomous AI coding agent · VS Code fork
Open-source AI code assistant
The world's most adopted AI developer tool
Terminal-based AI coding agent for large tasks
The fast, collaborative code editor with AI built in
Self-hosted AI coding assistant
Infrastructure
· 4 agentsFrameworks and scaffolding used to build agents. Not agents themselves · excluded from the capability leaderboard.
Multi-agent conversation framework
Framework for orchestrating role-playing AI agents
Build stateful, multi-actor agents with graphs
Tiny library for powerful agents
Frequently asked
Answers pulled from the agents dataset
What's the best AI coding agent right now?
OpenHands from All Hands AI leads the composite capability ranking with a score of 71.8 across SWE-bench Verified and other reasoning benchmarks.
What's the fastest-growing AI agent on GitHub?
LangGraph gained 603 GitHub stars in the last 7 days · ranked by velocity across the 25 agents tracked.
Can I run an AI coding agent on my laptop?
Yes · 16 of the 25 agents tracked are fully open-source. OpenHands, Aider, Cline, Roo Code, and Continue all run locally, bring-your-own-key, and work with open-weight models like DeepSeek and Qwen.
What powers Claude Code?
Claude Code is a terminal-first coding agent by Anthropic running on claude-opus-4-6 and claude-sonnet-4-6. It's a closed product with a ReAct loop and built-in file-edit, shell, and web-search tools.
What's the difference between an AI agent and an agent framework?
Agents are systems that do work · Claude Code, Devin, OpenHands. Frameworks like AutoGen, CrewAI, LangGraph are scaffolding you use to build agents. We track 4 frameworks in the Infrastructure tab, excluded from the capability leaderboard since frameworks don't have benchmark scores.
See also
Keep exploring the BenchGecko graph