PolyWorkBench (Claude Code): 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 Claude Code agent runtime, 1,800 s per task). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.8 | 92.3 |
| 2 | GLM-5.2 | 85.5 |
| 3 | GPT-5.5 | 81.5 |
| 4 | Qwen 3.6 27B | 80.1 |
| 5 | Claude Opus 4.7 | 79.7 |
| 6 | DeepSeek V4 Flash | 79.6 |
| 7 | Qwen 3.6 35B A3B | 79.3 |
| 8 | GLM-5.1 | 78.5 |
Interactive version: theaggregate.ai/benchmark?slug=polyworkbench-claude-code · How It Works · Data refreshed daily, snapshot 2026-09-29.