Pixtral-12B-2409: benchmark results
Mistral's Pixtral 12B (2409), a natively multimodal model pairing a 12B decoder with a 400M vision encoder, trained on interleaved image-text data with a 128K context window. Provider: Mistral. Released 2024-09-11. Access: Open.
Aggregate Rating 1474 ± 21, rank #2760 of 10916 rated models, from 190 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HELM Image2Struct - Image2Latex Wild | 71.86 | EMS | 100 |
| HELM Image2Struct - Image2Struct Wild | 87.5 | Mean win rate | 100 |
| HELM Image2Struct - Image2Webpage Wild | 74.75 | EMS | 100 |
| HELM VHELM - Gqa | 66.9 | PEM | 95.7 |
| HELM VHELM - Mmmu - Music | 46.67 | Prefix Quasi-Exact Match (%) | 95.7 |
| HELM VHELM - Blink Perception - Visual Similarity | 97.78 | Prefix Quasi-Exact Match (%) | 92.6 |
| HELM VHELM - Blink Perception - Art Style | 98.29 | Prefix Quasi-Exact Match (%) | 91.5 |
| HELM VHELM - Mmmu - Energy and Power | 53.33 | Prefix Quasi-Exact Match (%) | 84 |
| HELM VHELM - Hateful Memes | 66.8 | PEM | 80.9 |
| HELM VHELM - Toxicity | 80.85 | Mean win rate | 80.9 |
| HELM VHELM - Mmmu - Chemistry | 52.17 | Prefix Quasi-Exact Match (%) | 78.7 |
| HELM VHELM - Mmmu - Mechanical Engineering | 50 | Prefix Quasi-Exact Match (%) | 77.7 |
Interactive version: theaggregate.ai/model?slug=pixtral-12b-2409 · How It Works · Data refreshed daily, snapshot 2026-11-08.