ACU · 2026-11-06 14:26 UTC

Self-Hosted LLM on a Phone: The 9B Model Coinbase Fine-Tuned to Beat Opus

On October 12, Coinbase CEO Brian Armstrong said his engineers fine-tuned Qwen3.5-9B, a free open-weight model, on internal data and that it beat frontier models at fraud prevention. Coinbase used LoRA training on one Nvidia GPU for under $100, and the model beat its Opus 4.5 fraud reviewer on all four metrics, with 9.6 points higher F1 and a median answer in less than half the time. The article also reports Qwen3.5-9B running on 16 GB Android phones in the Acurast network for about 0.6 to 1 ACU per phone per day, with a peak of 9.3 tokens per second for writing on the fastest phones tested.

Loading the story on Tree News…