DuMateBench - Partial Pass Rate: leaderboard
The deterministic part of DuMateBench (2026): the share of each task checklist of explicit requirements that an agent harness and base model satisfy, averaged over 200 real-session workflow tasks, in percent; higher is better.
Metric: Partial pass rate (%; share of the task's deterministic checklist requirements that the produced artifacts satisfy, averaged over tasks; 200 tasks reconstructed from anonymized, privacy-screened DuMate production sessions, each in an isolated Docker container with insufficient, unstable and noisy environment conditions; each agent keeps its native control loop and runtime settings). Source: arxiv.org. Saturation forecast: Estimated already saturated. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 90.01 | #49 |
| 2 | Claude Opus 4.8 | 89.86 | #56 |
| 3 | DeepSeek V4 Pro | 87.21 | #188 |
| 4 | GLM-5.2 | 86.13 | #138 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=dumatebench-partial-pass-rate · How It Works · Data refreshed daily, snapshot 2026-11-13.