# Revolut Made a Self-Hosted Chinese Open-Weights Model Its Default for 2,000 Engineers and Says AI Coding Now Costs About a Tenth

Times of AI Desk · 2026-11-04 · Enterprise

[https://timesof.ai/2026/11/revolut-self-hosted-glm-default-coding-model-2000-engineers](https://timesof.ai/2026/11/revolut-self-hosted-glm-default-coding-model-2000-engineers)

> Revolut says it cut its AI coding inference costs by roughly 90% in under a year by making self-hosted Z.ai GLM models, now GLM-5.3 on 56 Nvidia B200 GPUs, the default for about 2,000 engineers instead of relying mainly on services like Claude Code, The Register reported. Its infrastructure lead said cheap tokens led engineers to hand debugging and ticket work to AI. Commercial models remain for vision and other exceptions.

Revolut's numbers make a concrete public case that a large company can swap frontier coding services for a self-hosted open-weights model and keep its engineers happy, at least for text-only work. The British fintech has cut its **AI coding inference costs by roughly 90%** in less than a year by making **self-hosted models its default** instead of relying mainly on commercial services like Anthropic's **Claude Code**, **The Register** reported on **November 3**.

## What Revolut runs

Speaking at the AI Engineer conference in New York last month, Revolut ops and infrastructure lead **David Abitbol** said the company rolled AI coding tools out to roughly **2,000 engineers**, The Register reported. Revolut started with **GLM-5.1**, from Chinese open-weights developer **Z.ai**, and now serves **GLM-5.3** on **56 Nvidia B200 GPUs**. At $0.30 per million tokens, self-hosting GLM-5.1 cost about a tenth of using OpenAI's Codex or Claude Code, according to the report.

Abitbol said the biggest gains came not from writing code but from letting models handle debugging and Jira and Confluence tickets once engineers stopped rationing tokens. "When you have unlimited tokens, you're free to do that," he said. He said Revolut aims for under three seconds to first token on contexts of about 256,000 tokens and about 200 tokens a second.

## Where frontier models still fit

GLM is text only, so Revolut routes vision queries to models such as Gemini or OpenAI's GPT Luna, The Register reported. "Open source is now a priority at Revolut, and our economics are favorable," Abitbol said. "The billing is flat, we don't have any bill shock at the end of the month, and the frontier is the exception." He said concerns about Chinese open-weights models apply mainly to using them through APIs run from China, not to running them on your own hardware.

Other companies are also rethinking coding-tool spend: [Replit moved most of its model spending from Anthropic to OpenAI](https://timesof.ai/2026/10/replit-shifts-ai-spend-openai-discount-anthropic) after a discount, and [Goldman Sachs limited Claude Code and GitHub to its engineers](https://timesof.ai/2026/11/goldman-sachs-limits-claude-code-github-to-engineers).

## Limits

- The cost, speed and productivity figures are Revolut's own, given in a conference talk and reported by one outlet; we have not seen the talk or any underlying data.
- The roughly 90% saving compares token costs and may not include the full cost of buying or renting and running the GPUs.
- Revolut has not said how its security team assessed the Chinese-developed model beyond Abitbol's remarks.

## Sources

- [The Register: British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hosted (November 3, 2026)](https://www.theregister.com/ai-and-ml/2026/11/03/british-fintech-revolut-slashes-ai-coding-costs-10x-by-cutting-out-claude-and-going-self-hosted/5307345)