The runtimes and harnesses people actually build agents with — ranked by live GitHub traction, alongside published SWE-bench results for agent scaffolds. OpenClaw vs Hermes Agent vs the coding-agent field, updated daily.
GitHub metrics retrieved 2026-09-12
| # | Framework | Category | Language | Stars | Forks | Last push | License | Source |
|---|---|---|---|---|---|---|---|---|
| 1 | OpenClaw The AI that really does things. Any OS. Any Platform. The lobster way. 🦞 |
Personal assistant | TypeScript | 389,475 | 81,872 | 2026-09-12 | Other | repo |
| 2 | Hermes Agent The agent that grows with you |
Personal assistant | Python | 244,671 | 50,729 | 2026-09-12 | MIT | repo |
| 3 | Claude Code Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands. |
Coding agent | Python | 144,799 | 23,117 | 2026-09-11 | — | repo |
| 4 | Codex CLI Lightweight coding agent that runs in your terminal |
Coding agent | Rust | 123,462 | 19,006 | 2026-09-12 | Apache-2.0 | repo |
| 5 | Gemini CLI An open-source AI agent that brings the power of Gemini directly into your terminal. |
Coding agent | TypeScript | 106,930 | 14,558 | 2026-09-12 | Apache-2.0 | repo |
| 6 | OpenHands 🙌 OpenHands: AI-Driven Development |
Coding agent | TypeScript | 87,577 | 11,469 | 2026-09-12 | MIT | repo |
| 7 | Cline Autonomous coding agent as an SDK, IDE extension, or CLI assistant. |
Coding agent | TypeScript | 67,863 | 7,331 | 2026-09-12 | Apache-2.0 | repo |
| 8 | AutoGen A programming framework for agentic AI |
Orchestration SDK | Python | 60,941 | 9,206 | 2026-04-15 | CC-BY-4.0 | repo |
| 9 | CrewAI Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks. |
Orchestration SDK | Python | 58,387 | 8,408 | 2026-09-11 | MIT | repo |
| 10 | aider aider is AI pair programming in your terminal |
Coding agent | Python | 48,913 | 4,945 | 2026-05-22 | Apache-2.0 | repo |
| 11 | LangGraph Build resilient agents. |
Orchestration SDK | Python | 41,492 | 7,011 | 2026-09-11 | MIT | repo |
| 12 | gcrab our project Small, safe, replayable self-hosted AI agent. Single Go binary, zero deps, hard 8K context budget, policy-gated exec, append-only event log. Telegram + CLI. |
Personal assistant | Go | 0 | 0 | 2026-09-10 | MIT | repo |
Source: GitHub REST API, live per-repo · retrieved 2026-09-12. Stars measure adoption/attention, not capability — see methodology.
gcrab is the token-frugal counterpoint to the big agent runtimes: a single static Go binary with zero dependencies that runs on a 512 MB VPS. A hard 8K context budget is enforced before every provider call (local-model friendly), every action lands in an append-only replayable event log, exec is policy-gated with an un-disableable deny backstop, and Telegram access is deny-by-default · 0 stars on GitHub.
Full disclosure: gcrab is built by the operators of this site. It receives no ranking boost anywhere on AgentLeaderboards — the table above sorts purely on public GitHub metrics, and gcrab's columns show exactly what the GitHub API returns.
gcrab.com →GitHub repo"Checked" means the SWE-bench maintainers verified the run themselves; "self-reported" entries were submitted with logs by the authors.
| # | Agent + model | Resolved | Date | Status |
|---|---|---|---|---|
| 1 | Sonar Foundation Agent + Claude 4.5 Opus | 79.2% | 2025-12-05 | self-reported |
| 2 | live-SWE-agent + Claude 4.5 Opus medium (20251101) | 79.2% | 2025-12-15 | self-reported |
| 3 | TRAE + Doubao-Seed-Code | 78.8% | 2025-09-28 | self-reported |
| 4 | live-SWE-agent + Gemini 3 Pro Preview (2025-11-18) | 77.4% | 2025-11-20 | self-reported |
| 5 | EPAM AI/Run Developer Agent v20250719 + Claude 4 Sonnet | 76.8% | 2025-08-04 | self-reported |
| 6 | Atlassian Rovo Dev (2025-09-02) | 76.8% | 2025-09-02 | self-reported |
| 7 | Claude 4.5 Opus (high) | 76.8% | 2026-02-17 | self-reported |
| 8 | ACoder | 76.4% | 2025-08-19 | self-reported |
| 9 | Gemini 3 Flash (high) | 75.8% | 2026-02-17 | self-reported |
| 10 | MiniMax M2.5 (high) | 75.8% | 2026-02-17 | self-reported |
| 11 | Warp | 75.6% | 2025-09-01 | self-reported |
| 12 | Claude 4.6 Opus | 75.6% | 2026-02-17 | self-reported |
| 13 | TRAE + Claude Sonnet 4 + Opus 4 + Sonnet 3.7 + Gemini 2.5 Pro | 75.2% | 2025-06-12 | self-reported |
| 14 | Harness AI | 74.8% | 2025-07-31 | self-reported |
| 15 | Sonar Foundation Agent + Claude 4.5 Sonnet | 74.8% | 2025-11-03 | self-reported |
| 16 | Lingxi-v1.5_claude-4-sonnet-20250514 | 74.6% | 2025-07-20 | self-reported |
| 17 | JoyCode + Claude 4 Sonnet + GPT-4.1 | 74.6% | 2025-09-15 | self-reported |
| 18 | Refact.ai Agent + Claude 4 Sonnet + o4-mini | 74.4% | 2025-06-03 | self-reported |
| 19 | Prometheus-v1.2.1 + GPT-5 | 74.4% | 2025-10-15 | self-reported |
| 20 | Claude 4.5 Opus (20251101) (medium) | 74.4% | 2025-11-24 | checked |
Source: swebench.com · fetched 2026-09-12, board last updated 2026-02-26 (198 days ago)
Full extracts: /api/v1/swebench.json