According to an X post by @guilleflorvs, around 2,000 Revolut engineers now code with the self-hosted model GLM by default, with Claude Code and Codex kept as the exception. The figures come from a public talk by David Abitbol, who leads AI operations at Revolut: 56 Nvidia B200s, about 30 cents per million tokens and roughly a tenth of the previous Anthropic and OpenAI bill, all self-reported. For SMEs the interesting part is less the model than the question behind it: once tokens are no longer rationed, teams hand more work to agents and need clear guardrails for that.
Source: @guilleflorvs on X ↗ via @guitaripod
Context: Single source on X, based on a public conference talk by Revolut. We found no separate official Revolut announcement.