An $8 ESP32-S3 now runs a 28.9M-parameter model fully offline
brianrkelly · x · 2026-07-24
A developer has pushed a 28.9M-parameter language model onto an ESP32-S3 microcontroller that costs about $8.
- It runs fully offline.
- It can generate coherent short stories at around 9.5 tokens/sec.
- The post says this is over 100× larger than earlier TinyStories-scale experiments on the same chip class.
- The image confirms the device is an ESP32-S3 board.
The broader point is how far “local AI” can be stretched on ultra-cheap hardware.
Related event: Developer Runs 28.9M Parameter Model Offline on $8 ESP32-S3(2 posts)→
More from Embodied
- OpenAI’s Codex keyboard ships, and early users say it’s pricey but fun — APPSO · 2026-07-24
- BFL launches FLUX 3 as a unified model for image, video, audio, and action prediction — umesh_ai · 2026-07-24
- TIME Features Unitree's GD01, the World's First Mass-Produced Transformable Mecha Robot — whurley · 2026-07-24
- OpenBMB opens up MiniCPM-Robot as its first embodied AI model family at WAIC — CyberRobooo · 2026-07-24
- Robot skins made by industrial knitting could enable scalable tactile sensing — mynkgoel · 2026-07-24
- Unitree shows Super Athlete AS2-W with 16 kg payload and 30+ km range — Scobleizer · 2026-07-24