Revolut moves to self-hosted GLM
Summary
Guillermo Flor reports that Revolut shifted about 2,000 engineers to self-hosted GLM for coding, with 56 Nvidia B200 GPUs and a token cost around 30 cents per million. Engineering lead time reportedly improved 70%, and gains appeared once token quotas were removed, with frontier models becoming the exception and open weights the default. The post highlights pricing and deployment implications for AI labs preparing for IPO.