Cohere Labs: data mixing lifts 3.35B model to 93%+ in-language reasoning across 60 languages
Cohere_Labs · x · 2026-09-11
Cohere Labs released "Building Multilingual Bridges," a data-centric study on enabling LLMs to reason in the user's language (L2 reasoning) instead of defaulting to English. Their 3.35B Tiny Aya L2-Thinker achieves >93% in-language reasoning across 60 languages on 6 benchmarks. Key findings: generalization to held-out languages requires broad language coverage, readily available multilingual non-reasoning data, and a strong English reasoning backbone.
More from Research
- DeepSeek's CED vs GLM's KV reuse: a developer unpacks how the cache-sharing modes actually differ — stochasticchasm · 2026-09-11
- SG-JEPA Paper: World Models That Train on Earth and Deploy on Mars, Halving Zero-Shot Physics Error — randall_balestr · 2026-09-11
- Code-as-Policy article explores general models learning to operate robots like software — yawnxyz · 2026-09-11
- ICML Generative AI and Creativity workshop releases new survey paper — lasha_nlp · 2026-09-11
- AVSplat: assist-view preconditioning fixes dense-view degradation in feed-forward 3DGS — zhenjun_zhao · 2026-09-11
- RRSI descriptor enables radiation, rotation and scale-invariant multimodal image matching — zhenjun_zhao · 2026-09-11