Tools & ProjectsNovember 15, 2026
2
š±
Revolut slashes AI coding costs by 90%
According to Revolut's David Abitbol, self-hosted GLM models cut AI coding costs by 90% for about 2,000 engineers.
#GLM#Self-Hosting#Inference Optimization#Open-Weight Models#Cost Reduction

š„ What happened
Revolut made self-hosted GLM models the default for 2,000 engineers and pushed Claude Code and Codex to the exception list. The bank cut blended token costs to 30 cents per million ā one-tenth of what it paid Anthropic and OpenAI.
š” Why it matters
56 Nvidia B200s serve the whole team because the mixture-of-experts model activates only 40B of 750B parameters per token. Revolut improved lead time by 70% not by buying more GPUs but by tuning inference and running Chinese weights inside EU-hosted compliance sandboxes.
ā” Our take
Frontier API pricing just lost its anchor: a regulated EU bank proved open weights plus constraint engineering is enough. Anyone still paying per-token to OpenAI in 2027 hasn't done the math.
The title, summary and analysis of this item were produced automatically by an AI system and have not been editorially reviewed. They may contain errors, bias or omissions ā when in doubt, read the linked original source.
