tiny-aya-base: benchmark results
Cohere Labs Tiny Aya Base is the 3.35B pretrained multilingual checkpoint for downstream adaptation, covering more than 70 languages. It remains separate from the Global, Fire, Earth and Water chat releases. Provider: Cohere. Released 2026-02-17. Access: Open.
Aggregate Rating 1403 ± 28, rank #5934 of 10916 rated models, from 264 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Bulgarian NLU - Cinexio | 63.23 | Matthews correlation coefficient (%) | 99.5 |
| EuroEval Lithuanian NLU - Atsiliepimai | 50.51 | Matthews correlation coefficient (%) | 99 |
| EuroEval Catalan NLU - Guia CAT | 76.04 | Matthews correlation coefficient (%) | 98.8 |
| EuroEval Finnish NLU - Tydiqa FI | 76.83 | Character F1 (%) | 97.6 |
| EuroEval Greek NLU - MultiWikiQA EL | 73.4 | Character F1 (%) | 97.5 |
| EuroEval German NLU - Germanquad | 68.95 | Character F1 (%) | 97.3 |
| EuroEval Portuguese NLU - SST-2 PT | 87.55 | Matthews correlation coefficient (%) | 96.6 |
| EuroEval Portuguese NLU - MultiWikiQA PT | 78.16 | Character F1 (%) | 96.3 |
| EuroEval Spanish NLU - MLQA ES | 69.3 | Character F1 (%) | 95.8 |
| EuroEval English NLU - SQuAD | 87.78 | Character F1 (%) | 95.4 |
| EuroEval Estonian NLU - MultiWikiQA ET | 73.59 | Character F1 (%) | 94.8 |
| EuroEval Italian NLU - SQuAD IT | 74.68 | Character F1 (%) | 94.8 |
Interactive version: theaggregate.ai/model?slug=tiny-aya-base · How It Works · Data refreshed daily, snapshot 2026-11-08.