Revolut moves to self-hosted GLM
The gist
Revolut moved about 2,000 engineers off Claude Code and Codex onto self-hosted GLM models, cutting AI coding costs roughly 90%.
The move tests whether cheap open-weight models can replace frontier tools at a regulated fintech.
How it unfolded
- Apr 2026Revolut moved to self-hosted GLM 5.1, which its own chart credits with a 70% gain in lead time and 60% in coding time.
- Apr 2026 to Aug 2026Revolut held its GPU count fixed and tuned inference parameters instead of buying more hardware, cutting time-to-first-token from over 10 seconds to about 1.6 seconds — an 88% performance gain against 76% higher demand.
- Oct 13, 2026David Abitbol detailed the move at the AI Engineer NYC conference: about 2,000 engineers shifted off Anthropic's Claude Code and OpenAI's Codex onto self-hosted GLM models, cutting AI coding costs by about 90%.