# Thoughts on Revolut switching default coding AI from Claude Code / Codex to self-hosted GLM-5.3? — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/9bd9dcefa020995e

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-11-03 23:29 ingested 2026-11-04 02:40

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1xn8owx/thoughts_on_revolut_switching_default_coding_ai/)

## retained text



```
Just saw the news that Revolut has made self-hosted open-weight models (GLM-5.1 → now GLM-5.3 from Z.ai / Zhipu) the default coding assistant for \~2,000 engineers.  
Key points from David Abitbol’s talk (AI Engineer conference):  
•  Previously used Claude Code (and Codex) with strict budgets because some engineers were burning $1k–$10k/month with limited ROI.  
•  Switched to self-hosted GLM running on 56× Nvidia B200 GPUs.  
•  Blended cost ≈ $0.30 per million tokens → roughly 1/10th of what they were paying Anthropic/OpenAI (≈90% cost cut).  
•  Measured \~60% improvement in coding time and \~70% improvement in lead time.  
•  Unlimited tokens → engineers now use AI for Jira, Confluence, Slack, docs, etc., not just “hard” coding tasks.  
•  Claude Code / Codex / Gemini are still available but treated as the exception (mainly for complex or vision tasks). GLM is text-only so they route images elsewhere.

Curious what people here think:  
1.  Is this the start of a bigger shift where open/self-hosted models become the default and frontier APIs are only for the hard 10–20% of work?  
2.  Anyone already running similar setups (GLM, DeepSeek, Qwen, etc.) in production for coding agents? How does quality compare day-to-day vs Claude Code / Cursor / Codex?  
3.  For a regulated fintech, how big a deal is the compliance/self-hosting angle vs pure cost?  
4.  Does this put real pressure on Anthropic/OpenAI pricing for enterprise coding use cases?

Would love to hear experiences, skepticism, or “we tried this and…” stories.
```



document fetched 2026-11-04 00:39 · canonical https://reddit.com/r/ClaudeCode/comments/1xn8owx/thoughts_on_revolut_switching_default_coding_ai
