Evo-Bench - Search: 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 128 BrowseComp and 128 HLE held-out tasks; one run). Source: arxiv.org. Saturation forecast: Around May 2028. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.8 (Max) | 46.5 |
| 2 | GLM-5.2 (Max) | 45.4 |
| 3 | GPT-5.6 Sol (Max) | 44.5 |
| 4 | DeepSeek V4 Pro (Max) | 34.4 |
| 5 | Gemma 4 31B (IT) (Thinking) | 24.2 |
Interactive version: theaggregate.ai/benchmark?slug=evo-bench-search · How It Works · Data refreshed daily, snapshot 2026-09-26.