Get the latest tech news
Smollm3: Smol, multilingual, long-context reasoner LLM
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
It includes architecture details, exact data mixtures showing how we progressively boost performance across domains in a three-stage pretraining approach, and the methodology for building a hybrid reasoning model. The chat template serves as the interface that enables seamless switching between reasoning and non-reasoning modes, and its design directly impacts both our training data format and model behavior. While Qwen3 4B generally achieves the highest scores across both thinking and non-thinking modes, SmolLM3 demonstrates competitive performance within the 3B parameter class, particularly excelling in mathematical reasoning and complex problem-solving tasks.
Or read this on Hacker News