Woofun AI reports that Coinbase CEO Brian Armstrong announced a near 50% reduction in AI expenditure alongside rising token usage. The firm achieved this by optimizing default settings, routing, and caching strategies. Specific actions include switching defaults to open-weight models such as GLM 5.2 and Kimi 2.7, noting that 91% of staff had not hit usage limits. The company also implemented custom prompt preprocessing for task-specific routing, boosted the LibreChat cache hit rate from 5% to 60%, and streamlined context management by initiating new sessions for different tasks.

Additionally, engineers retain model selection freedom but must account for the associated cost impacts.