Revolut cut AI coding costs 90% by ditching Claude for self-hosted models
Revolut spent months rationing AI access for its engineers because commercial API bills ran $1,000 to $10,000 per month. The fix was not a better contract. The British fintech moved its 2,000-engineer team off Anthropic’s Claude Code and OpenAI’s Codex and onto self-hosted, open-weight models running on its own hardware.
The Numbers
David Abitbol, ops and infrastructure lead at Revolut, shared the details at the AI Engineer conference in New York. After the rollout, Revolut measured a 60% improvement in coding time and a 70% improvement in lead time, which tracks how long work moves from start to delivery. Earlier models, including Gemini, had produced only 10 to 20% productivity gains, which helped push the case for self-hosting.
The models running today come from Z.ai, a Chinese AI lab. Revolut deployed GLM-5.1 in April 2026 and later moved to GLM-5.3. That newer model runs across 56 Nvidia B200 GPUs inside Revolut’s own infrastructure. Blended inference cost: approximately $0.30 per million tokens.
What Changed After Go-Live
Once token rationing was gone, engineers started handing off the work they had avoided before: writing documentation, filing Jira tickets, the surrounding chores that nobody wants to do but that eat real hours.
Revolut also tuned performance between April and August 2026. Throughput passed 7 requests per second, a 76% increase. Time-to-first-token dropped about 88%, settling at 1.6 to 1.68 seconds.
The Compliance Piece
Revolut is a regulated financial company. The setup keeps AI weights on US infrastructure and keeps humans reviewing code suggestions. With an open-weight model, proprietary code does not have to leave the company’s own environment, which matters for a bank. Self-hosting required upfront hardware investment, infrastructure talent, and months of optimization work, as the April to August tuning period shows. At 2,000 engineers, the economics justify all of it.
The Operator Takeaway
This is a scale story, not a small team playbook. But the underlying principle transfers: token rationing is a productivity tax, and if your AI bills are forcing your team to think twice before using the tool, the cost-per-use math deserves a hard look. Self-hosting has an entry price. Revolut’s data suggests the exit price of staying on commercial APIs is higher.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
