WeaveBench - Design: leaderboard
WeaveBench (2026) tests computer-use agents on 114 long-horizon real-world tasks that need GUI, command-line, code and API tools together; this series comes from the paper table in which each model drives a fixed OpenClaw agent runtime at its best thinking mode. This board is the pass rate (%) on the design tasks; higher is better.
Metric: PassRate (%) on the Design domain tasks of 114 long-horizon real-world computer-use tasks that need GUI, CLI, code and API tools together, each model run as the backbone of a fixed OpenClaw agent runtime with the full hybrid tool pool at its best thinking mode; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (High) | 40 | #49 (GPT-5.5) |
| 2 | GPT-5.4 (High) | 20 | #123 (GPT-5.4) |
| 3 | Claude Opus 4.7 (High) | 20 | #79 (Claude Opus 4.7) |
| 4 | GPT-5.3 Codex (High) | 20 | #116 (GPT-5.3 Codex) |
| 5 | Gemini 3.1 Pro (Preview) | 0 | #94 |
| 6 | Qwen 3.5 397B A17B | 0 | #246 |
| 7 | Qwen 3 VL 8B (Thinking) | 0 | #887 |
| 8 | GPT-5.1 Codex (High) | 0 | #288 (GPT-5.1 Codex) |
| 9 | GPT-5.2 Codex (High) | 0 | #189 (GPT-5.2 Codex) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=weavebench-design · How It Works · Data refreshed daily, snapshot 2026-11-13.