Coinbase Switches to Chinese AI Models GLM and Kimi, Cuts AI Spending by 50%
Coinbase CEO Brian Armstrong disclosed Friday that the cryptocurrency exchange has switched its default AI models to GLM 5.2 from Chinese lab Zhipu and Kimi 2.7 from Beijing-based Moonshot AI, cutting the company's internal AI spending by nearly half even as token consumption hits record highs [1]. The shift, announced in a post on X outlining five cost-reduction levers, marks one of the most public enterprise defections from Western AI providers to Chinese alternatives.
Coinbase has made a significant shift in its AI infrastructure by defaulting engineers to GLM 5.2 from Zhipu and Kimi 2.7 from Moonshot AI through its internal LLM gateway. This change has resulted in nearly a 50% reduction in AI spending.
The GLM 5.2 model offers substantial cost savings, priced at $1.40 per million input tokens compared to $5 for Anthropic’s Opus 4.8 — roughly five times cheaper. Despite the lower cost, it outperforms on the SWE-bench Pro coding benchmark.
Coinbase also implemented an automated routing system and caching improvements. These changes boosted the cache hit rate from 5% to 60%, representing a 12x improvement in efficiency.
Other U.S. companies, including Snowflake and AI startup Lindy, have similarly begun adopting Chinese open-weight models to optimize their AI expenses and performance.
Meanwhile, Coinbase shares have declined 33% year-to-date, currently trading at roughly $151 with a market capitalization near $40 billion.
Reader Support
Support WebNovare
Send a crypto tip if this article helped you.
Log in or register to like, bookmark, watch tags, and comment.
Comments
No comments yet.
Log in or register to join the conversation.