Revolut Moved About 2,000 Engineers Onto Self Hosted GLM for Coding, Making Claude Code and Codex the Exception, at About a Tenth of the Cost
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list. Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%. Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything. Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO. h/t @OliverMolander Would you move your team off frontier APIs if the bill dropped 90%?
Context
A big regulated company making open weight models its default for coding is a direct pricing threat to frontier labs that sell tokens through APIs.
Revolut is a large European fintech company with a banking license. GLM is a family of open weight models from the Chinese lab Z.ai (Zhipu), which companies can run on their own hardware instead of paying per token through an API. The figures come from a talk by David Abitbol, who leads AI operations at Revolut, listed on the AI Engineer NYC 2026 site.
Deep dive
Guillermo Flor (@guilleflorvs) posted on Nov 2 that Revolut moved about 2,000 engineers onto self hosted GLM for coding and pushed Claude Code and Codex to the exception list. The post, which credits Oliver Molander for the tip, passed 290,000 views.
According to the post, the figures come from David Abitbol, who leads AI operations at Revolut. He is quoted as saying the company runs GLM on 56 Nvidia B200 GPUs at a blended cost of 30 cents per million tokens, roughly a tenth of what Revolut paid Anthropic and OpenAI, and that engineering lead time improved 70%. Other posts citing the same talk name the model as GLM 5.3. These numbers are Revolut's own account as relayed on X; they have not been independently checked.
Flor argued that the gains showed up once tokens stopped being rationed, because engineers stopped saving AI for the hardest tickets and used it on everything. He framed the shift as frontier models becoming the exception and open weights the default, and called that a pricing problem for labs heading toward an IPO.
The AI Engineer NYC 2026 site lists the talk as Shipping like a startup, regulated like a bank, describing Revolut's move toward self hosted coding models, the inference tuning that made them practical, and the controls used in a regulated engineering organization. The story lands the same day this deck covers Anthropic's amended IPO filing and its revenue numbers.
What to expect next
Watch whether other large companies publish similar numbers on moving coding work to self hosted open weight models, and whether Anthropic or OpenAI respond with lower enterprise prices or private deployment options. Open questions include how the 30 cents per million tokens figure was calculated, whether it counts hardware purchase and staff costs, and which tasks Revolut still sends to frontier models. Further details may appear when the full talk recording circulates.