Claw-Eval: leaderboard
Agentic task-completion benchmark for complex multi-step scenarios, evaluating whether models can plan, act, and recover in realistic workflows.
Metric: Score (%). Source: claw-eval.github.io. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 81.4 |
| 2 | MiMo-V2.5-Pro | 80.6 |
| 3 | Step 3.7 Flash | 80.5 |
| 4 | Claude Opus 4.6 | 80.4 |
| 5 | GPT-5.4 | 78.4 |
| 6 | MiMo-V2.5 | 78.4 |
| 7 | Kimi K2.6 | 77.9 |
| 8 | Gemini 3.1 Pro (Preview) | 77.3 |
| 9 | GLM-5.1 | 77.3 |
| 10 | DeepSeek V4 Pro | 77.2 |
| 11 | MiMo-V2-Pro | 77 |
| 12 | Muse Spark | 76.3 |
| 13 | Qwen 3.6 Plus | 75 |
| 14 | Kimi K2.5 | 74.9 |
| 15 | GLM-5-Turbo | 74.4 |
Interactive version: theaggregate.ai/benchmark?slug=claw-eval-hx1n6f · How It Works · Data refreshed daily, snapshot 2026-10-09.