Revolut made Zhipu GLM the default AI coding model for about 2,000 engineers, while Claude Code and Codex became backup tools for complex tasks, according to a Beating AI News Flash report cited by BlockBeats.
Reported cost shift
Revolut AI operations head David Abitbol said the company previously set a $300-per-person monthly quota for Claude Code. Since April, it has self-deployed GLM and offered unlimited tokens. The company now uses about 56 NVIDIA B200 GPUs to run GLM 5.3 at a blended cost of roughly $0.30 per million tokens, described as one-tenth of earlier Anthropic and OpenAI spending.
Internal performance figures
Internal company statistics cited in the report showed coding-time metrics improving about 60% and assignment-to-delivery cycle time improving about 70%. Engineers also expanded AI use to internal documents, Jira tickets and Slack requirements.
The cost and performance figures are attributed to Revolut’s executive and internal statistics; the flash does not provide an independent audit or a forecast. Source: BlockBeats English report, published Nov. 3, 2026.