Coinbase fine-tunes Qwen3.5-9B with internal data, its anti-fraud performance outperforms leading large models.

Coinbase CEO Brian Armstrong announced on social media that the company fine-tuned the open-source model Qwen3.5-9B using its proprietary internal data, resulting in a model that outperforms leading large language models (LLMs) specifically on anti-fraud tasks, while being lower-cost and faster. Relevant charts show the post-trained model significantly outperforms both the unadapted version and the Opus 4.5 decision agent on metrics including accuracy, recall, F1 score, and dollar-weighted recal

@tonmelonfest