Revolut switches 2000 engineers' default coding model to self-hosted GLM

Revolut, Europe's most valuable startup at roughly a $200 billion valuation, has gone public with a major shift in its enterprise AI strategy. Earlier this year, around 2000 engineers were using Claude Code and Codex on usage-based billing, burning $1,000–$10,000 per person per month with returns that didn't justify the cost. The company then switched its default coding model to a self-hosted Chinese open-weight model, GLM, and moved frontier APIs like Claude Code and Codex to an exceptions list. The news was revealed by David Abitbol, Revolut's head of AI operations, in a talk at AI Engineer NYC.

Confirmed

Default model switched to self-hosted GLM; Claude Code and Codex demoted to exception options
Infrastructure consists of about 56 Nvidia B200 GPUs (rented from Together AI), serving all 2000 engineers
TTFT (time to first token) cut from 10 seconds to 1.6 seconds
Coding bill dropped to roughly 1/10 of previous levels; some accounts claim 60% savings in coding time, others say engineering delivery sped up 70% — refer to Abitbol's talk for exact figures

Why it matters

Many commentators see this as a landmark case of a large enterprise abandoning frontier closed-source APIs in favor of an open-weight model, showing that in high-frequency internal workloads like coding, self-hosted open models can genuinely rival frontier models on both cost and performance
Its cost ledger ($1,000–$10,000 per user per month vs. 1/10 after self-hosting) offers a quantified benchmark for other companies considering a similar switch

2026-11-02 ~ 2026-11-03 · 5 related posts

Primary sources

Revolut's self-hosted GLM playbook: 56 B200s, TTFT cut from 10s to 1.6s, 1/10 the cost — AccBalanced · 2026-11-03 11:28 (UTC+08:00)
Revolut cuts AI costs 10x by self-hosting GLM on 56 B200s, boosting lead time 70% — SumitGup · 2026-11-02 07:02 (UTC+08:00)
Revolut ditches Claude/Codex for open-source GLM, cuts AI bill by 90% — 2C_ornot2C · 2026-11-03 10:00 (UTC+08:00)

[source] Revolut cuts AI costs 10x by self-hosting GLM on 56 B200s, boosting lead time 70% — SumitGup · 2026-11-02
Revolut switched coding default to open-source GLM, cutting bills to 1/10 and coding time 60% — 2C_ornot2C · 2026-11-03
Revolut moves ~2,000 engineers to self-hosted GLM, cutting coding costs ~90% — AccBalanced · 2026-11-03
[source] Revolut's self-hosted GLM playbook: 56 B200s, TTFT cut from 10s to 1.6s, 1/10 the cost — AccBalanced · 2026-11-03

1 near-duplicate retellings: 2C_ornot2C