SmolLM3-3B

SmolLM3-3B is a 3-billion-parameter open-weight language model from Hugging Face's Smol models research line, designed for reasoning, multilingual use across six languages, and long-context tasks up to 128k tokens. It emphasizes full openness — releasing weights and training details — alongside a hybrid reasoning mode and instruct tuning trained on an 11.2-trillion-token curriculum. Source: supabase.