Coinbase fine-tunes Qwen3.5-9B model for fraud prevention
Summary
Coinbase has successfully fine-tuned an open weight model, Qwen3.5-9B, using proprietary internal data for fraud prevention, achieving superior performance over general frontier models at lower cost and speed. This development reflects a growing trend in AI specialization, where firms are adapting models to efficiently tackle specific tasks. Such specialized models are expected to play a significant role in evolving financial operations, potentially replacing traditional backend services as organizations work towards advancements in agentic finance.
Analysis
Coinbase: Coinbase is a cryptocurrency exchange and digital asset platform focused on enabling trading, custody, and related financial services. The company recently fine-tuned the open-weight Qwen3.5-9B model on its proprietary internal data for the specialized task of fraud prevention, where it outperformed general frontier models while delivering advantages in cost and speed. Coinbase is also developing internal tools and harnesses to let employees easily fine-tune and serve such models for tasks like payments risk review. Qwen3.5-9B: Qwen3.5-9B is an open-weight large language model in the Qwen series designed for broad general capabilities. Coinbase selected this model as the base for internal fine-tuning on proprietary data, using it to build a more efficient specialist system for fraud prevention that surpassed frontier models on the narrow task. AI Specialization: Companies are increasingly fine-tuning open-weight models on internal data to create specialist systems that handle narrow tasks more effectively than general frontier models. Agentic Finance Development: Organizations are exploring architectures where fine-tuned models could eventually replace certain backend services in financial operations.
Categories
aitechai_agentsmachine_learningvirtuals