PUMA Polish Multimodal - Structured Extraction: leaderboard
Structured extraction category of PUMA (Polish Unified Multimodal Assessment), a culturally grounded Polish benchmark for vision-language models (Documents group; arXiv 2608.21853v1). Strict score, the percentage of answers meeting every verification rule, 0-100, higher is better, from the OPI-PIB/puma Space data.json; audio-pipeline rows and the auxiliary Soft score are excluded.
Metric: Strict Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 80 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 79 | #64 |
| 2 | Muse Wonder | 75 | #18 |
| 3 | Gemini 3.6 Flash | 71 | #57 |
| 4 | Claude Fable 5 | 70 | #20 |
| 5 | Gemini 3.7 Flash | 70 | #24 |
| 6 | Grok 4.7 | 70 | #26 |
| 7 | Gemini 4 Argon | 69 | #9 |
| 8 | Gemini 2.5 Pro | 68 | #159 |
| 9 | Gemini 3 Flash | 68 | #103 |
| 10 | Gemini 3.8 Flash | 68 | #30 |
| 11 | Gemini 3.5 Flash | 66 | #66 |
| 12 | Muse Spark 1.2 | 66 | #69 |
| 13 | Claude Sonnet 5.5 | 65 | #14 |
| 14 | Claude Opus 5.5 | 62 | #3 |
| 15 | Claude Sonnet 5 | 60 | #65 |
No result here: #10 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=puma-polish-multimodal-structured-extraction · How It Works · Data refreshed daily, snapshot 2026-11-05.