Cactus Releases Needle 2: A 14MB On-Device Agent Model
Cactus has introduced Needle 2, an open-source micro agent model designed for edge devices. Compressed to a mere 14MB, the model requires only 28MB of RAM and achieves over 500 tokens/s decoding speed on a Raspberry Pi 5.
2026-08-11 ~ 2026-08-11 · 2 related posts
- Needle 2: A 14MB Agentic LLM Hitting 500 Tokens/sec on Raspberry Pi 5 — Henrie_the_dreamer · 2026-08-11
- Needle 2: A 14MB Agentic LLM for Tiny Devices Released — julianharris · 2026-08-11