A developer has successfully deployed a 28.9 million parameter language model on an ESP32 microcontroller, costing just $8. The implementation demonstrates practical AI inference on ultra-low-cost hardware.
The project, shared on GitHub as esp32-ai, shows that capable language models can run on severely resource-constrained devices. The ESP32 microcontroller, widely available and inexpensive, typically features 4MB of flash storage and limited RAM.
Running LLMs on microcontrollers addresses edge computing use cases where cloud connectivity is unavailable or undesirable. Inference happens locally without network latency or privacy concerns associated with cloud services.
The implementation generated 122 points and 27 comments on Hacker News, indicating significant interest from the developer community. This work follows broader trends in model optimization, including quantization and pruning techniques that reduce model size without sacrificing substantial performance.
The approach opens possibilities for embedding AI capabilities directly into IoT devices, industrial equipment, and embedded systems where computational power and cost are primary constraints.
Sony has launched the ULT Tower, a new party speaker that supports Auracast technology for wireless multi-speaker connectivity. The speakers can be chained together for synchronized audio across multiple units.
Dell has launched the 14S, a budget laptop positioned as a competitor to Apple's MacBook Neo. The device appears to offer solid performance at a lower price point.
GrapheneOS has confirmed that Google's Pixel 11 includes Memory Tagging Extension (MTE) support, reversing earlier expectations that the feature would be absent from the device.