According to CoinMeta, Brian Armstrong, the CEO of Coinbase, stated that Coinbase utilizes internal data to fine-tune the open-source model Qwen3.5-9B. This approach results in better performance in anti-fraud tasks compared to cutting-edge models, with lower costs and faster speeds. The model from Coinbase has a 9.60% higher score than leading models, and the end-to-end latency is reduced by 55.00%. Armstrong believes that dedicated models will prevail in specialized tasks, and Coinbase is currently enabling employees to easily fine-tune models internally; some back-end services will be replaced by this new architecture in the future.