IndustryBench-MIPU: leaderboard
IndustryBench-MIPU (2026) tests multi-image industrial product understanding: from all images of a product (specification tables, nameplates, technical drawings) the model extracts product-level property-value pairs, scored against multi-model consensus annotations of 4,559 products. Product-level F1 in percent; higher is better.
Metric: Product-level F1 (%) of multi-image extraction (the model receives all valid images of a product and outputs product-level property-value pairs), structured attribute value extraction from industrial product images (specification tables, nameplates, technical drawings): property names matched exactly and values by rule-based normalization then a Qwen 3.6 Plus semantic judge; full extraction prompt, thinking enabled where supported; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 65.1 | #94 |
| 2 | Qwen 3.5 397B A17B | 62.7 | #246 |
| 3 | GPT-5.4 (Thinking) | 60.5 | #123 (GPT-5.4) |
| 4 | Qwen 3.5 Plus (Thinking) | 59.9 | #232 (Qwen 3.5 Plus) |
| 5 | Claude Opus 4.6 (Thinking) | 57.2 | #102 (Claude Opus 4.6) |
| 6 | Kimi K2.5 (Thinking) | 56.7 | #256 (Kimi K2.5) |
| 7 | Qwen 3.5 27B | 55.8 | #359 |
| 8 | Qwen 3.5 122B A10B | 50.1 | #331 |
| 9 | Qwen 3.5 35B A3B | 20.6 | #482 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=industrybench-mipu · How It Works · Data refreshed daily, snapshot 2026-11-13.