[Coinbase Switches Default AI Models to Open-Source GLM5.2, Kimi2.7, Cutting AI Costs by Half] BlockBeats News, June 28: Coinbase CEO Brian Armstrong disclosed that the company has reduced its AI expenses by nearly half by switching its default AI models to open-source models such as GLM 5.2 and Kimi 2.7, combined with measures like intelligent model routing and cache optimization, even as token usage continues to grow exponentially. Brian Armstrong stated that Coinbase no longer relies on usage limits or spending alerts to control costs but instead defaults to more cost-effective open-source models, only invoking more advanced, high-performance models in necessary scenarios. Additionally, the company has further reduced inference costs through automatic model routing, improving cache hit rates, and streamlining context. It is worth noting that this statement comes at a time when Anthropic's CEO has recently expressed concerns multiple times about open-source models and model distillation. Coinbase's approach demonstrates that in enterprise-level AI applications, open-source models are becoming an important choice for balancing cost and efficiency. [Original Link]