IndustryCode - Sub-Problems: leaderboard
Metric: Pass@1 (%) on the sub-problems of the IndustryCode test split (106 main problems decomposed into 499 sub-problems; industrial Python, C++, MATLAB and Stata projects with test cases), unweighted mean of the Python, C++, MATLAB and Stata sub-problem pass rates; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 68.1 |
| 2 | Claude Sonnet 4.5 | 64.4 |
| 3 | Gemini 3 Pro | 63.4 |
| 4 | Claude Haiku 4.5 | 57.4 |
| 5 | Qwen 3 Max | 55.9 |
| 6 | Grok 4.1 | 53.6 |
| 7 | GPT-5.2 | 53.4 |
| 8 | Gemini 2.5 Pro | 51.6 |
| 9 | Qwen 3 Max (Thinking) | 49.3 |
| 10 | GPT-5.1 | 48.1 |
| 11 | GLM-4.7 (Thinking) | 47 |
| 12 | Kimi K2 | 46.8 |
| 13 | GLM-4.7 (Non-reasoning) | 40.8 |
| 14 | Kimi K2 (Thinking) | 37.6 |
| 15 | Qwen 3 Coder Plus | 36.1 |
Interactive version: theaggregate.ai/benchmark?slug=industrycode-sub-problems · How It Works · Data refreshed daily, snapshot 2026-10-07.