ComBench: leaderboard
Metric: Overall average normalized score (%) over four sampled solutions across the 100 problems (proof rubric score for analysis items, verifier-gated score for construction items); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 65.4 |
| 2 | Gemini 3.1 Pro (Preview) (Thinking) | 60.3 |
| 3 | Kimi K2.6 | 53.5 |
| 4 | DeepSeek V4 Pro | 45.2 |
| 5 | Qwen 3.6 Max Preview | 24.9 |
| 6 | GLM-5.1 | 23.6 |
| 7 | Qwen 3.6 35B A3B | 20.3 |
| 8 | Nemotron Cascade 2 30B A3B | 19.6 |
| 9 | Gemma 4 31B (IT) | 16.8 |
Interactive version: theaggregate.ai/benchmark?slug=combench · How It Works · Data refreshed daily, snapshot 2026-09-29.