IndustryCode - Sub-Problems: leaderboard

IndustryCode (2026) is a code-generation benchmark of hierarchical industrial programming problems in Python, C++, MATLAB and Stata, each main problem decomposed into sub-problems checked by test cases. This board is Pass@1 in percent on the sub-problems, the unweighted mean over the four languages; higher is better.

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 February 2027. 17 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.568.1#160
2Claude Sonnet 4.564.4#261
3Gemini 3 Pro63.4#145
4Claude Haiku 4.557.4#583
5Qwen 3 Max55.9#415
6Grok 4.153.6#462
7GPT-5.253.4#185
8Gemini 2.5 Pro51.6#276
9Seed 1.8Thinking50.1#252
10GPT-5.148.1#240
11GLM-4.7Thinking47#380
12Kimi K246.8#443
13Kimi K2 (Thinking)37.6
14Qwen 3 Coder Plus36.1#720
15DeepSeek V3.2 Exp35.3#439

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=industrycode-sub-problems · How It Works · Data refreshed daily, snapshot 2026-11-21.