Guillermo Flor says Revolut is migrating about 2,000 engineers to self-hosted GLM

Nov 3

1. 19:37

   Analytics India Magazine

   Revolut self-hosts GLM models, cutting AI coding inference costs by about 90%

   UK fintech company Revolut has reduced the inference cost of AI coding tasks to about one-tenth of what it was when using Anthropic and OpenAI by self-hosting Z.ai's open-weight GLM models, with a blended cost of 30 cents per million tokens. Its AI business lead David Abitbol said at the AI Engineering conference that the deployment uses 56 NVIDIA B200 GPUs to serve about 2,000 engineers running GLM-5.3, shortening engineering delivery cycles by 70% and reducing coding time by 60%. Revolut still retains Claude Code, Codex and Gemini for specialized tasks such as image recognition, while the internally hosted GLM handles most of the coding work.

   Expand summary
2. 06:11

   Hacker News Trending

   Revolut Migrates About 2,000 Engineers to Self-Hosted GLM for Coding, Claude Code and Codex Become Exceptions

   According to Guillermo Flor, relaying statements from Revolut AI operations lead David Abitbol, Revolut has migrated the coding work of about 2,000 engineers to self-hosted GLM and placed Claude Code and Codex on an exception list. Its configuration is 56 Nvidia B200 GPUs, with a blended cost of about 30 cents per million tokens, roughly one-tenth of what it previously paid Anthropic and OpenAI, and engineering delivery cycles improved by 70%. The author believes that once tokens were no longer rate-limited, engineers applied AI to all tasks, and that frontier models are becoming the exception while open weights become the default choice.

   Expand summary