PolyWorkBench (OpenClaw): leaderboard
Metric: Mean Grade (%; weighted deterministic rubric checks of the output artefacts, mean over 67 multilingual long-horizon workplace tasks in five domains; best of up to three runs, Pass@1; the OpenClaw ReAct-style scaffold with shell, file and Python tools, 1,800 s per task). Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GLM-5.2 | 85.3 |
| 2 | GPT-5.5 | 79.4 |
| 3 | GLM-5.1 | 78.3 |
| 4 | Qwen 3.6 27B | 78.2 |
| 5 | Claude Opus 4.8 | 77.8 |
| 6 | Claude Opus 4.7 | 70.9 |
| 7 | DeepSeek V4 Flash | 70.8 |
| 8 | Qwen 3.6 35B A3B | 67.3 |
Interactive version: theaggregate.ai/benchmark?slug=polyworkbench-openclaw · How It Works · Data refreshed daily, snapshot 2026-09-29.