What's this about?
A report that Coinbase moved significant AI workloads to open Chinese models — GLM and Kimi — and halved its AI spending. It's a concrete corporate data point landing right under the week's open-weights policy fight.
What exactly is the trade-off here?
Open Chinese models are near-frontier and far cheaper to run (you or a provider host them), versus closed US APIs that are pricier but carry brand, support, and — the policy argument goes — different data-governance and geopolitical baggage. Coinbase's move says that for its workloads, the price gap outweighed those concerns.
Why does it matter?
Policy debates are abstract; procurement decisions are real. A major US public company halving its bill by adopting open Chinese models is exactly the dynamic Anthropic's position paper is worried about — and exactly the competitiveness argument the open-weights letter's signatories make.