ClawBench: leaderboard
ClawBench evaluates AI browser agents on live-web, multi-step everyday tasks such as booking, ordering, applications, email, and shopping. The input modality is text and web interaction. Its pinned metric is Reward (Lenient) (%), based on whether the final request matches the task schema and whether an LLM judge deems the payload to satisfy the instruction. Source: BenchmarkList’s attributed result table.
Metric: Reward (Lenient) (%). Source: benchmarklist.com. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | claude-opus-4-6 | 61.4 | |
| 2 | claude-sonnet-4-6 | 56.9 | |
| 3 | claude-opus-4-7 | 44.6 | |
| 4 | gpt-5.5 | 35.4 | |
| 5 | glm-5.1 | 34.6 | |
| 6 | deepseek-v4-pro | 33.9 | |
| 7 | claude-haiku-4-5-20251001 | 30.1 | |
| 8 | gpt-5.4-2026-03-05 | 25.5 | |
| 9 | gpt-5.4-mini-2026-03-17 | 24.8 | |
| 10 | kimi-k2.5 | 17.6 |
Interactive version: theaggregate.ai/benchmark?slug=clawbench · How It Works · Data refreshed daily, snapshot 2026-11-05.