EdgeBench - Math: leaderboard
Metric: Formal mathematics tasks: mean best-so-far score (0-100) after a 12-hour continuous agent run with evaluator feedback, three runs per task, averaged over the family's tasks; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 5 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.8 | 55 |
| 2 | GPT-5.5 | 50 |
| 3 | GPT-5.4 | 40.8 |
| 4 | GLM-5.1 | 24.6 |
| 5 | DeepSeek V4 Pro | 14.1 |
Interactive version: theaggregate.ai/benchmark?slug=edgebench-math · How It Works · Data refreshed daily, snapshot 2026-09-29.