IndustryCode: leaderboard
Metric: Pass@1 (%) on the main 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 main-problem pass rates; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 42.5 |
| 2 | Gemini 3 Pro | 41.2 |
| 3 | Claude Sonnet 4.5 | 33.8 |
| 4 | Qwen 3 Max | 32.5 |
| 5 | GPT-5.2 | 32.4 |
| 6 | Claude Haiku 4.5 | 31.8 |
| 7 | Grok 4.1 | 31.3 |
| 8 | GLM-4.7 (Thinking) | 28.6 |
| 9 | Qwen 3 Max (Thinking) | 28.4 |
| 10 | Gemini 2.5 Pro | 23 |
| 11 | DeepSeek V3.2 Exp | 22.4 |
| 12 | Qwen 3 Coder Plus | 21.6 |
| 13 | Kimi K2 | 20.8 |
| 14 | GPT-5.1 | 18.9 |
| 15 | GLM-4.7 (Non-reasoning) | 14.9 |
Interactive version: theaggregate.ai/benchmark?slug=industrycode · How It Works · Data refreshed daily, snapshot 2026-10-07.