Brian Armstrong said Coinbase is routing large language model usage through its own gateway, defaulting to GLM 5.2 and Kimi 2.7 as token consumption continues to rise.
Coinbase is testing a new artificial intelligence setup that Brian Armstrong said has reduced spending by half even as token usage keeps increasing. The Coinbase CEO said the company is using its own LLM (large language model) gateway and is defaulting to open-weight models including GLM 5.2 and Kimi 2.7. The post frames the move as a cost-efficiency shift in AI infrastructure, where companies route demand through an internal model layer to manage usage and lower inference expenses while maintaining growing workloads.