PANews June 27 news, Coinbase CEO Brian Armstrong stated on X platform that through infrastructure optimization (default settings, routing and caching mechanisms), the company reduced AI spending by nearly half amid exponential growth in token usage. The measures include:

• Better default models: Default to open-source/cost-effective models (such as GLM 5.2, Kimi 2.7). 91% of employees originally never hit usage limits.
• Intelligent routing: Automatically route tasks to the most suitable model (frontier models for planning, cheaper models for execution), letting AI choose instead of humans.
• Efficient caching: Significantly increase cache hit rate (e.g., LibreChat from 5% to 60%).
• Streamlined context: Start new sessions for new tasks, narrow file scope, disconnect unused tools, reduce waste.
• Greater visibility: Allow free usage but make spending visible.