ClawBench v1: leaderboard
ClawBench v1 (TIGER-Lab) — agentic computer-use tasks in the Claw environment; reward rate across scripted episodes.
Metric: Reward Rate (%). Source: huggingface.co. Saturation forecast: Around January 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 61.4 | #63 |
| 2 | Claude Sonnet 4.6 | 56.9 | #90 |
| 3 | Claude Haiku 4.5 (20251001) | 30.1 | #261 |
| 4 | GPT-5.4 (2026-03-05) | 25.5 | #59 |
| 5 | GPT-5.4 Mini (2026-03-17) | 24.8 | #169 |
| 6 | Kimi K2.5 | 17.6 | #144 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=clawbench-v1 · How It Works · Data refreshed daily, snapshot 2026-10-20.