In detail
- Coinbase now uses Chinese models instead of OpenAI/Anthropic; 91% of developers no longer exceed their old usage limits.
- Automatic routing system selects the best model per request based on task, price, and caching potential; caching optimization raised hit rate from 5 to 60%.
- Lindy CEO and Snowflake also testing Chinese models; OpenAI and Anthropic face pricing pressure—OpenAI offers GPT-5.6-Sol with better token efficiency.
- Coinbase ties spending to expected business impact: 'The more you spend on AI, the more impact we expect.'
Why it matters
Established companies switching to Chinese models signals a market inflection: cost optimization becomes a competitive lever, and Western labs must rethink pricing. For SMEs, this means concrete alternatives to expensive US APIs.
For you Compare total cost of ownership (token price + caching efficiency) of DeepSeek, Kimi, and GLM against your current OpenAI/Anthropic spend—savings could be substantial.