Key Notes
- Revolut says its GLM 5.3 deployment serves about 2,000 engineers on 56 Nvidia B200 GPUs at a blended cost near $0.30 per million tokens.
- Its AI operations leader reported spending around one-tenth of the previous Anthropic and OpenAI bill, while retaining frontier models for exceptions.
- Lower costs have let engineers use AI more freely for Jira, documentation and other work around coding.
Revolut has made a self-hosted Chinese AI model the default for about 2,000 engineers, cutting its bill to roughly one-tenth of what it previously spent on Anthropic and OpenAI, according to the executive leading its AI operations. The company runs GLM 5.3 on about 56 Nvidia B200 GPUs at a blended cost near $0.30 per million tokens.
David Abitbol described the deployment in an October 13 presentation at AI Engineer NYC. Revolut began self-hosting GLM in April and now keeps frontier tools such as Claude Code and Codex available for exceptions. The change gives engineers a cheaper default for routine work while preserving access to other models when a task calls for them.
A Token Budget Changes Which Tasks Get Done
Revolut previously imposed strict budgets on Claude Code. With lower serving costs, engineers have more room to use AI for tasks around the code itself, including searching their inboxes, handling Jira tickets and finding material in Confluence. Abitbol’s account puts those activities at the center of the productivity argument. Faster code generation helps, but engineers also spend time locating requirements and working out what needs to change.
A spending cap encourages people to save their allowance for work they consider difficult enough to justify it. A lower internal cost changes that calculation. An engineer can ask an agent to find a specification or summarize a discussion without treating every request as a decision about how much of the month’s budget to use. The benefit depends on whether that work saves time and produces a result the engineer can use.
The Hardware Still Needs to Earn Its Keep
The $0.30 figure is a blended cost reported for Revolut’s deployment. The cost of a dedicated cluster depends on how much work it serves, how efficiently the software runs and how much capacity sits idle. Those factors make utilization as important to the economics as the choice of model. A company with a different volume or pattern of requests could face a different cost per token.
Revolut’s operating target is to begin responding in under three seconds, Abitbol said. Complaints start arriving once the wait exceeds five seconds. His team tuned the serving system to improve response times on the existing GPUs. For an engineer repeatedly handing work to an agent, that delay can determine whether the tool fits comfortably into the working day.
Z.ai distributes GLM 5.3’s model weights, allowing organizations to run it on infrastructure they select. That gives a company control over deployment and serving, along with responsibility for operating the system. Revolut’s use of an open-weight model is one example of how businesses can choose different models and hosting arrangements for different workloads.
Revolut Is Building More Than a Coding Assistant
The engineering deployment sits alongside a broader investment in AI. In August, Revolut announced a dedicated research division and described PRAGMA, its proprietary foundation model developed with Nvidia for financial risk, operations and recommendations. GLM’s role in assisting engineers addresses a different part of that effort.
Model costs are becoming a practical constraint as agents take on more work. Elon Musk recently cited serving costs as Grok Bot users reported reaching their weekly limits. Revolut’s approach moves more of that cost decision to the team running the infrastructure. Engineers get a lower-cost default, and the company manages the capacity behind it.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
