WeaveBench: leaderboard
WeaveBench pass rate over 114 long-horizon computer-use tasks, with agent harnesses kept separate. Published one-decimal measurements are preserved; source-supported High effort and Gemini Preview checkpoint settings remain explicit. The unsuffixed board uses OpenClaw. Board: WeaveBench.
Metric: PassRate (%). Source: weavebench.github.io. Saturation forecast: Around March 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 (High) | 35.1 | #49 (Claude Opus 4.7) |
| 2 | GPT-5.5 (High) | 33.3 | #31 (GPT-5.5) |
| 3 | GPT-5.4 (High) | 22.8 | #79 (GPT-5.4) |
| 4 | GPT-5.3 Codex (High) | 18.4 | #69 (GPT-5.3 Codex) |
| 5 | GPT-5.2 Codex (High) | 6.1 | #95 (GPT-5.2 Codex) |
| 6 | Gemini 3.1 Pro (Preview) | 1.8 | #58 |
| 7 | GPT-5.1 Codex (High) | 1.8 | #159 (GPT-5.1 Codex) |
| 8 | Qwen 3.5 397B A17B | 0.9 | #146 |
| 9 | Qwen 3 VL 8B (Thinking) | 0.9 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=weavebench · How It Works · Data refreshed daily, snapshot 2026-10-20.