ComBench - Construction Pass Rate: leaderboard
Metric: Per-sample construction pass rate (%): the share of construction-centric solutions whose witness is accepted by the deterministic verifier; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 75 |
| 2 | Kimi K2.6 | 75 |
| 3 | Gemini 3.1 Pro (Preview) (Thinking) | 75 |
| 4 | DeepSeek V4 Pro | 62.5 |
| 5 | Qwen 3.6 35B A3B | 40.5 |
| 6 | Qwen 3.6 Max Preview | 38 |
| 7 | GLM-5.1 | 27 |
| 8 | Nemotron Cascade 2 30B A3B | 20.5 |
| 9 | Gemma 4 31B (IT) | 15 |
Interactive version: theaggregate.ai/benchmark?slug=combench-construction-pass-rate · How It Works · Data refreshed daily, snapshot 2026-09-29.