Evo-Bench: leaderboard
Metric: Held-out score (%; the fixed DeepSeek-V4-Flash policy runs the harness the model evolved from a CodeAct seed within 20 iterations, 1,000 steps and 48 hours; mean of the five source benchmarks (BrowseComp, HLE, GDPval, APEX-Agents, Claw-Eval) over the 448-task evaluation suite; one run). Source: arxiv.org. Saturation forecast: Around 2029. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 46.3 |
| 2 | Claude Opus 4.8 (Max) | 45.8 |
| 3 | GLM-5.2 (Max) | 43.5 |
| 4 | DeepSeek V4 Pro (Max) | 39.1 |
| 5 | Gemma 4 31B (IT) (Thinking) | 35.9 |
Interactive version: theaggregate.ai/benchmark?slug=evo-bench · How It Works · Data refreshed daily, snapshot 2026-09-26.