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

#ModelScore
1Claude Opus 4.892.3
2GLM-5.285.5
3GPT-5.581.5
4Qwen 3.6 27B80.1
5Claude Opus 4.779.7
6DeepSeek V4 Flash79.6
7Qwen 3.6 35B A3B79.3
8GLM-5.178.5

Interactive version: theaggregate.ai/benchmark?slug=polyworkbench-claude-code · How It Works · Data refreshed daily, snapshot 2026-09-29.