Cohere releases Tiny Aya L2-Thinker: 3.35B model with native reasoning in 45 languages
davlanade · x · 2026-09-11
Cohere for AI researchers unveiled Tiny Aya L2-Thinker, a 3.35B-parameter model with 32K context trained for in-language reasoning across 45 languages.
- Their data-centric recipe mixes multilingual reasoning, non-reasoning, and English reasoning data, achieving over 90% in-language reasoning rate across 60 languages on 6 benchmarks spanning math, open-ended generation, cultural/common-sense reasoning, and instruction following.
- They release open weights plus multilingual reasoning data covering 44 languages beyond English—the largest such dataset they're aware of.
- Motivation: reasoning is central to modern models, yet the window into model thinking is still English-only; the team asks whether reasoning transfers across languages.
More from Models
- Developer: Astra is a good model — jxnlco · 2026-09-11
- New Model Hits Opus-Level Benchmarks at Wild Efficiency, RL Infra Details Emerge — nrehiew_ · 2026-09-11
- V4.1's Reasoning Curve Isn't Linear; Agent Teams Explicitly RL-Trained to Collaborate — nrehiew_ · 2026-09-11
- V4.1 Is the First Model With a Numerical Reasoning Effort Parameter Tied to Length Penalty — nrehiew_ · 2026-09-11
- Only 15 Fused Kernels in Prefill: V4.1's Inference Stack and Minute-Scale SWA Cache — nrehiew_ · 2026-09-11
- DeepSeek V4.1 Flash post-training is fully data-centric: synthetic multi-agent trajectories plus checkpoint merging — nrehiew_ · 2026-09-11