Needle 2

Needle 2 is an open 45-million-parameter LLM designed for tool calling, device interaction, and structured information extraction on extremely resource-constrained hardware. Distributed as a single 14 MB binary requiring 28 MB of RAM, it runs efficiently on devices without GPUs or NPUs such as microcontrollers (e.g., ESP32-S3), budget smartphones, Raspberry Pis, wearables, and smart home devices. The model supports running locally for privacy-sensitive, low-latency AI applications, offering production-ready capabilities with fine-tuning options. It features built-in confidence scoring for offloading to the cloud when needed and employs novel compression and architectural techniques to minimize computation and memory usage while maintaining competitive accuracy on function calling benchmarks.