AI model runs on $10 microcontroller using Google's embedding technique

By PulseAugur Editorial · [1 sources] · 2026-07-27 13:07

A Ukrainian developer named Slava S, also known as 'slvDev', has successfully run a 28.9-million-parameter AI model on a low-cost ESP32-S3 microcontroller, which retails for around $10. This was achieved by employing Google's Per-Layer Embeddings technique, adapted from the Gemma model, to manage the model's memory requirements. The embedding table was stored in the slower 16MB Flash memory, while the core reasoning weights utilized the limited 512KB of SRAM, enabling the model to operate entirely on-device. AI

IMPACT Demonstrates the potential for running sophisticated AI models on inexpensive, embedded devices, opening possibilities for offline AI applications.

RANK_REASON Demonstration of running a large language model on extremely low-cost, resource-constrained hardware. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Tom's Hardware →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

AI model runs on $10 microcontroller using Google's embedding technique

COVERAGE [1]

Tom's Hardware TIER_1 English(EN) · Zak Killian · 2026-07-27 13:07

AI developer runs 28.9-million-parameter model on $10 ESP32-S3 microcontroller — uses Google's Per-Layer Embeddings technique, stores table on 16MB Flash memory

Getting a local language model running on a sub-$10 microcontroller is impressive despite its obvious limitations.

COVERAGE [1]

AI developer runs 28.9-million-parameter model on $10 ESP32-S3 microcontroller — uses Google's Per-Layer Embeddings technique, stores table on 16MB Flash memory

RELATED ENTITIES

RELATED TOPICS