OWE-Bench - ML: leaderboard
OWE-Bench - ML — Expert-built long-horizon science, software, ML and engineering tasks with task-specific hardware and4–18hour budgets; native baseline/ceiling normalized held-out reward,mean of8attempts and equal task means. Web frozen October1,2026;deterministic sealed verifier. Native model/effort and every attempt retained. Score fraction0–1;higher is better.
Metric: Mean normalized held-out task score (fraction, 0–1). Source: openworldengineer.mercor.com. Saturation forecast: Around September 2027. 14 models tracked. Updated 15 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Claude Fable 5.2 | Max | 0.5456 | #1 |
| 2 | Claude Sonnet 5.5 | Max | 0.4778 | #19 |
| 3 | Claude Opus 5.5 | Max | 0.4258 | #5 |
| 4 | GPT-6.1 Sol | Max | 0.3935 | #10 |
| 5 | GPT-6 Astra | Max | 0.3894 | #6 |
| 6 | Gemini 4 Carbon | High | 0.3639 | #2 |
| 7 | Gemini 4 Argon | High | 0.3215 | #12 |
| 8 | Claude Fable 5.1 | Max | 0.3187 | #15 |
| 9 | DeepSeek V4.1 Flash | Max | 0.1679 | #62 |
| 10 | GLM-5.3 | Max | 0.1571 | #68 |
| 11 | Muse Wonder | Max | 0.1194 | #28 |
| 12 | Grok 4.7 | xHigh | 0.06599 | #47 |
| 13 | Beam 1.0 | Max | 0.02765 | #206 |
| 14 | Inkling | Max | 0.007109 | #281 |
Interactive version: theaggregate.ai/benchmark?slug=owe-bench-ml · How It Works · Data refreshed daily, snapshot 2026-11-16.