Blog

We introduce Luth-2-0.8B and Luth-2-2B, two small language models reaching state-of-the-art French performance, post-trained on Qwen3.5 with SFT, RL and multi-domain on-policy distillation.

We introduce two compact, non-reasoning causal LLMs, instruction-tuned entirely on French data, achieving state-of-the-art results for models of this size on several French benchmarks.