needle: 26m function call model that runs on incredibly small devices

Needle is a tiny AI model with just 26 million parameters, designed to run function calls on small consumer devices like phones, watches, and glasses. Distilled from Gemini, it processes up to 6,000 tokens per second and outperforms several larger models on single-shot function calling tasks. Its weights are fully open, and users can fine-tune it locally on a standard Mac or PC.