aya-vision-8B: benchmark results
Cohere Aya Vision 8B multimodal model for multilingual vision-language tasks. Provider: Cohere. Released 2025-03-04. Access: Open.
Aggregate Rating 1428 ± 12, rank #1107 of 1656 rated models, from 98 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Cedar - Text-Only - Japanese | 37.8 | Accuracy (%) | 86.7 |
| LatamBoard - Spanish TeleIA | 71.43 | Score (%) | 86.4 |
| LatamBoard - Spanish MGSM | 12 | Score (%) | 72.7 |
| LatamBoard - FaQuAD NLI | 74.19 | Score (%) | 69.7 |
| MUNIChus - French | 14.71 | CIDEr of the zero-shot French news image caption generated f | 66.7 |
| EVALITA - word-in-context | 63.86 | CPS | 61.7 |
| JMMMU-Pro - Japanese Heritage | 30.67 | Accuracy (%) | 61.5 |
| LatamBoard - ASSIN2 RTE | 91.06 | Score (%) | 60.6 |
| LatamBoard - Portuguese Score | 85.93 | Score (%) | 60.6 |
| KOFFVQA - Recognition | 75 | Score (%) | 58 |
| NormViz-Bench | 15.4 | Group Accuracy (%) | 55 |
| LatamBoard | 63.2 | LatamBoard Category Average (percentage) | 53.8 |
Interactive version: theaggregate.ai/model?slug=aya-vision-8b · How It Works · Data refreshed daily, snapshot 2026-10-20.