'Needle' debuts as a lightweight model distilled from Gemini's tool-calling capabilities to run on smartphones, with developers touting its usefulness for building mobile AI agents

Cactus Compute announced Needle, a 26M-parameter function-calling model distilled from Gemini-3.1-Flash-Lite.

The company says Needle runs locally on smartphones and other small devices with high speed, making it useful for mobile AI agents.

Needle is released under an MIT license and is available on Hugging Face, alongside Cactus Chat demos.

The launch highlights the promise of low-latency on-device AI, while also raising questions about Gemini distillation restrictions.