Aya 101
Cohere's 101-language instruction-tuned open model — the multilingual breadth milestone.
About
Aya 101, released by Cohere For AI in February 2024 (preprint February 13, 2024), was the first open model to chase multilingual breadth at scale: a 13B-parameter mT5 model instruction-tuned on 101 languages, over half of them lower-resourced. It was built on the Aya Collection — 513 million prompt/completion instances assembled with 3,000+ collaborators across 119 countries — plus xP3x and data-provenance mixtures.
The release paired the model with a new multilingual evaluation suite covering 99 languages across discriminative and generative tasks, and it substantially outperformed prior massively multilingual open models. Cohere was candid about the trade-off: spreading capacity across 101 languages meant per-language quality lagged, which motivated the narrower, deeper Aya 23 follow-up. The model card lists an Apache 2.0 license; as with any community mirror, confirm terms on the official Hugging Face page (CohereForAI/aya-101).
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 80; Trending score 66/100 from researcher assessment.