What is the new self-hosted GLM architecture at Revolut?
Revolut has transitioned approximately 2,000 engineers to a self-hosted GLM for coding tasks, relegating frontier APIs to an exception list according to NVIDIA Lanza una Herramienta de Auto-Ontología Open Source para Desarrolladores in the broader ecosystem. David Abitbol, who leads AI operations at the company, detailed the new deployment.
The infrastructure relies on 56 Nvidia B200 accelerators running locally.
Why did Revolut shift away from external frontier APIs?
The financial impact of the migration has drastically reduced operational overhead for artificial intelligence tasks. Costs fell to a blended thirty cents per million tokens, roughly a tenth of what the company previously paid Anthropic and OpenAI.
This reduction in costs addresses challenges similar to those seen in El colapso bursátil de Intel empeora con caídas récord en chips regarding hardware economics.
How did removing token limits affect engineering productivity?
Eliminating token rationing completely changed how developers interact with artificial intelligence tools on a daily basis. Engineers stopped saving AI queries strictly for hard tickets and began pointing the model at everything.
As a result of removing these usage constraints, engineering lead time improved by 70 percent across the board.