ClawArena: leaderboard
ClawArena evaluates coding agents in 12 evolving multi-session scenarios over 337 rounds, using multiple-choice and execution checks on workspace files and chat histories. The official board reports Composite Reliability Score (0–100) under each model’s primary framework: Claude Code for Anthropic models and OpenClaw for the others; higher is better.
Metric: Composite Reliability Score. Source: www.clawarena.cc. Saturation forecast: Around June 2027. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 68.28 | #31 |
| 2 | Gemma 4 31B | 63.8 | #200 |
| 3 | GPT-5.1 | 63.28 | #135 |
| 4 | GLM-5.1 | 60.63 | #97 |
| 5 | MiMo-V2.5-Pro | 59.65 | #158 |
| 6 | GPT-5.4 | 58.99 | #79 |
| 7 | Gemini 3.1 Pro (Preview) | 57.59 | #58 |
| 8 | Kimi K2.5 | 57.24 | #144 |
| 9 | Qwen 3.6 27B | 55.85 | #170 |
| 10 | DeepSeek V4 Pro | 55.25 | #102 |
| 11 | Qwen 3.6 Plus | 55.17 | #120 |
| 12 | GPT-5.2 | 54.18 | #110 |
| 13 | Qwen 3.6 35B A3B | 48.86 | #226 |
| 14 | GLM-4.7 Flash | 38.82 | #509 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=clawarena · How It Works · Data refreshed daily, snapshot 2026-10-20.