Cohere details Tiny Aya L2-Thinker: data mixing unlocks reasoning in 60 languages
Cohere_Labs · x · 2026-10-06
Cohere Labs published a blog explaining Tiny Aya L2-Thinker, a multilingual model that reasons in the user's language across 60 languages. The key idea is data mixing: curating training data proportions so models don't just answer in the native language but actually reason in it. The post walks through why in-language reasoning matters and how to unlock it.
Related event: Cohere Launches Tiny Aya Family of Multilingual Small Models(2 posts)→
More from Models
- No word yet on ML4's license, and the community is noticing — xeophon · 2026-10-06
- Mistral AI claimed to be state-of-the-art on GDPR compliance — onetwoval · 2026-10-06
- Why models never give up: RL training punishes quitting, so no chill version exists — rickasaurus · 2026-10-06
- Mistral Large 4 trails GLM-5.3 on Artificial Analysis index, fueling 'eval crisis' talk — Yuchenj_UW · 2026-10-06
- Mistral-Large-4.0 weights (1T total, 52B active) expected in ~25 days — paf1138 · 2026-10-06
- Mistral Large 4 pricing: $1.13 per task, 4x costier than similar open weights models — ArtificialAnlys · 2026-10-06