August 10, 2026
Small model, huge comment-section energy
Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots
Tiny AI for cheap gadgets drops — and the comments instantly split between wow and "is this just a random button presser"
TLDR: Needle 2 is a tiny AI designed to run on cheap gadgets without the cloud, which could make smart features far more accessible. Commenters were split between impressed excitement and roast-mode skepticism, especially after a demo moment where a simple input somehow turned into "lock the front door."
A startup just unveiled Needle 2, a shockingly tiny AI model meant to run directly on cheap phones, watches, smart-home gear, little robots, and even hobby devices like a Raspberry Pi. The big brag is simple: instead of needing a pricey computer or sending everything to the cloud, this thing is supposed to work right on the device, fast, private, and with very little memory. In plain English, it’s aiming to be the brains behind everyday actions like turning on lights, using apps, or pulling neat structured info out of text — all from hardware most people would normally call underpowered.
But the real fireworks came from the crowd. One camp was openly impressed, calling it "really cool" and treating it like a glimpse of the long-promised future where affordable gadgets finally get useful AI. Another immediately wandered into comedy: one user typed "HN" into the demo and got a command to lock the front door, which is the kind of accidental chaos that instantly turns a launch into meme fuel. That led to the spiciest divide of the thread: is this a clever little helper for simple device tasks, or, as one blunt commenter sneered, basically a "random sentence generator" wearing a lab coat? Others were more practical, asking how this would fit with voice control and wake words on a screenless device. So yes, the launch landed — but the comments made it entertaining: equal parts optimism, skepticism, and "please don’t let my toaster freestyle."
Key Points
- •The article announces Needle 2, an open 45M-parameter model for tool calling, device control, and structured extraction, packaged as a 14MB binary with about 28MB peak session RAM.
- •Needle 2 is positioned for low-cost edge hardware, including budget phones, Raspberry Pi devices, microcontrollers, wearables, robots, smart-home devices, and automotive systems.
- •Benchmark claims in the article say Needle 2 trades wins with FunctionGemma 270M, LFM2.5 230M, and Apple FM while being 5x to 70x smaller and using 2-bit precision.
- •Reported performance figures include 500 tokens/sec on Raspberry Pi 5, 400-1,500 tokens/sec on Meta Quest 3S and Apple Vision Pro, and 300-700 tokens/sec on Samsung A-Series phones.
- •The article says Needle 2 uses schema-constrained function calling, confidence-based cloud escalation, end-to-end 2-bit training with Cactus Quants, and a dependency-free C++ runtime that can also be fine-tuned locally on a Mac or PC.