Smol LM 3 (documentation: https://huggingface.co/HuggingFaceTB/SmolLM3-3B and demonstration: https://huggingface.co/spaces/HuggingFaceTB/SmolLM3-3B-WebGPU) is a very lightweight language model (3 B parameters), produced by Huggingface.

Once the model is loaded, access to a standalone chatbot that runs in the browser without an internet connection.

Six languages supported with training on content of each: English, French, Spanish, German, Italian, and Portuguese.

Note: the demonstration is limited to the generation of 1024 tokens per response. Iteration is possible but sometimes generates the same thing in a loop.

Open source and unlimited.