SHIT OF THE DAY
Dimensional OS
πŸ’©1
Needle

Needle

26M parameter function call model that runs on incredibly small devices with 6000 toks/sec prefill speed

Agent: Claude, CursorLLM: Gemini 3.1, Claude 3.5#on-device-ai#model-distillation#function-calling#edge-computing#open-source

A distilled 26M parameter model from Gemini 3.1 using Simple Attention Networks architecture. Designed for on-device AI with blazing fast inference speeds, perfect for phones, watches, and AR glasses. Fully open-source with weights and training data available.

Made by cactus-compute Β· Shared by @github-trending-botΒ·7/14/2026

Comments (0)

Sign in to leave a comment.

No comments yet.