--- title: Revolut moves about 2,000 engineers to self-hosted GLM... description: Revolut moved about 2,000 engineers onto a self-hosted GLM model for coding, making Claude Code and Codex the exception rather than the default. The setup runs... canonical: https://daily.dev/posts/revolut-moves-about-2-000-engineers-to-self-hosted-glm-for-coding-xv3uxpqtr twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Revolut moves about 2,000 engineers to self-hosted GLM for coding | daily.dev og:description: Revolut moved about 2,000 engineers onto a self-hosted GLM model for coding, making Claude Code and Codex the exception rather than the default. The setup runs... og:url: https://daily.dev/posts/revolut-moves-about-2-000-engineers-to-self-hosted-glm-for-coding-xv3uxpqtr og:image: https://api.daily.dev/og/posts/xV3UxpqTr.png og:image:alt: Revolut moves about 2,000 engineers to self-hosted GLM for coding og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Revolut moves about 2,000 engineers to self-hosted GLM for coding **[Collections](https://daily.dev/sources/collections)** · 2 min read · 2 upvotes · 0 comments ## Summary Revolut moved about 2,000 engineers onto a self-hosted GLM model for coding, making Claude Code and Codex the exception rather than the default. The setup runs on 56 Nvidia B200 GPUs at a blended cost of about 30 cents per million tokens, roughly a tenth of what the company paid Anthropic and OpenAI. David Abitbol, who leads AI ops at Revolut, reports engineering lead time improved by 70%. The improvement appeared once tokens stopped being rationed and engineers used AI on every ticket instead of saving it for hard ones. ## Content Revolut has moved roughly 2,000 engineers onto a self-hosted GLM model for coding. Claude Code and Codex are now on an exception list rather than the default. The numbers come from David Abitbol, who leads AI operations at Revolut, as relayed by @guilleflorvs: - The setup runs on 56 Nvidia B200 GPUs. - The blended cost is about 30 cents per million tokens, roughly a tenth of what the company paid Anthropic and OpenAI. - Engineering lead time improved 70%. ## Why the gains showed up The interesting part isn't the price. It's what happened once tokens stopped being rationed. Engineers no longer saved AI for the hard tickets and pointed it at everything. That's the claimed source of the lead-time improvement, and it seems plausible to me, though the 70% figure is Revolut's own and nobody outside has checked it. The pattern is frontier models as the exception and open weights as the default. If it spreads, that's a pricing problem for every lab heading toward an IPO. @omarsar0 sees it as part of a wider trend: companies that own their intelligence stack stay competitive, which he says is the role open-source AI plays now. His framing is to think of companies as mini-AGIs. ## What it doesn't show Revolut is large, and this probably doesn't translate to a five-person startup with no GPU hardware or DevOps team. Those teams will keep paying for Anthropic and OpenAI APIs. Two other caveats: - GLM may not match the frontier models Revolut used before. One commenter said the difference is negligible for seasoned engineers, but that's an opinion, not a benchmark. - Running 56 B200s takes a team to keep them healthy. The blended cost per token may already account for this, but the staffing burden is real and eats into the savings. The open question is the one @guilleflorvs asked: would you move your team off frontier APIs if the bill dropped 90%? For a company at Revolut's scale, apparently yes. ## Questions this post answers ### How much does Revolut pay per million tokens running a self-hosted GLM model for coding? Revolut's blended cost is about 30 cents per million tokens, running on 56 Nvidia B200 GPUs. That is roughly a tenth of what the company previously paid Anthropic and OpenAI for Claude Code and Codex, which are now used as exceptions rather than the default for its roughly 2,000 engineers. _Teams weighing self-hosted versus frontier AI coding tools can track real cost comparisons like this on daily.dev._ ### Why did Revolut see a 70% improvement in engineering lead time after switching AI coding tools? The improvement came after tokens stopped being rationed, not from the new model itself. Engineers previously saved AI usage for hard tickets, but once the self-hosted GLM setup made tokens cheap, they started using it on every ticket, which is what produced the 70% engineering lead time gain according to Revolut's AI ops lead David Abitbol. _Developers rethinking when and how much to use AI assistants can follow cases like this on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 1 discussion and 151 comments across hackernews (as of 2026-11-03). **TL;DR:** Discussion centers less on Revolut's move itself and more on whether cheaper/self-hosted open models (GLM, Qwen) can really replace frontier models like Opus, alongside a long tangent about whether engineers should personally invest in learning AI tools; cost/hardware economics of self-hosting also get heavily scrutinized. **Sentiment:** 30% positive · 35% mixed · 35% skeptical **The case for** - Several report that cheap local or open models (Qwen, GLM) are already good enough for most day-to-day coding work when paired with a good harness. - Removing token rationing and giving engineers unlimited/flatrate access is seen as the real productivity unlock, not just the specific model used. - Self-hosting avoids data-privacy/compliance concerns tied to sending code to third-party US cloud providers. **The pushback** - Many argue frontier models like Opus are still noticeably better at reasoning, following intent, and avoiding flailing compared to GLM/Qwen alternatives. - Skepticism that the hardware investment math holds up long-term given rapid GPU depreciation and shifting vendor pricing. - Zero-data-retention promises from cloud/API providers are seen as weak 'trust me' contractual claims rather than real guarantees. **By community** - hackernews (mixed): Opinions split heavily between those satisfied with cheaper open models and those insisting frontier models remain meaningfully better, with a long side debate about career incentives and job security overshadowing the core story. **Hottest debate:** Whether open/self-hosted models like GLM or Qwen are genuinely comparable to frontier models like Opus for serious coding work, or just a budget compromise. **Open questions** - Will this self-hosting trend generalize beyond companies with strict compliance/regulatory needs? - How sustainable is the hardware investment given GPU depreciation and rapidly evolving model generations? - Will frontier labs respond with cheaper high-volume enterprise pricing to compete with self-hosted setups? **Highlights** > If you were to take away my work's $500 budget for Opus 5.5 and replace it with unlimited use of GLM-5.3, it would be a huge downgrade in my opinion. If your employer is so cheap as to spend a $100 or less per month per engineer, while being large enough not to have access to the subscription plans, a move to self hosted or cheap providers does make sense in my view. > — [user43928 on hackernews · 6 comments](https://news.ycombinator.com/item?id=50350618) > > If you were to take away my work's $500 budget for Opus 5.5 and replace it with unlimited use of GLM-5.3, it would be a huge downgrade in my opinion. I am perfectly able to do the tasks with own custom harness and unlimited local Qwen 3.8 27B with 150t/s. Harness needs search capability and more fine-tuning for the job. You need less words to describe the intent with Astra/Opus 5.5, and it automatically usually observes the best practice that you don't need to explicitly tell, but with only a slight more effort you can get the job done. I have even made security audits with Qwen 3.8 27B against projects fully written with Claude Opus and it is a bit mind-boggling, but there have been tens of vulnerabilities, identified by this model alone. This excludes the issues found when guiding more myself. > — [nicce on hackernews](https://news.ycombinator.com/item?id=50351176) > I've used extensively everything between Opus 4.6, GPT-5.5, Fable 5, Opus 5, GPT-5.6 Sol, GPT-6 Astra, 6.1 Sol and Opus 5.5. To a lesser degree also Sonnet and Luna. And when I say extensively I mean 500h+ on a side project in addition to my full-time job. Even the difference between 5.6 Sol, which is miles ahead of local Qwen 3.8 27B, and Opus 5.5 is immediately obvious in the quality of the results. Having been stuck on 5.6 Sol at my job while Opus 5.5 was available at home, I can say with full confidence that the difference is significant, even if I weren't looking at the output at all. The way it understands my requests, forms a hypothesis, verifies the result and finishes the change is night and day different. This happens while 5.6 Sol is still flailing around somewhere in the monorepo struggling with even locating the relevant code. > — [user43928 on hackernews · 1 comments](https://news.ycombinator.com/item?id=50352100) > > 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. So they've invested around $2.5M-3M into hardware + monthly operating costs (electricity, updates, cooling). I wonder how much did they spend on tokens to decide to make such a switch. I mean if they used to spend $500k-1M per month just on tokens then it would pay off in a year or so, assuming their infra will be stable and efficient. We'll see if that's a trend for mid-big companies or just an experiment. All that math is assuming that it's all self-hosted in their data center or so and not paying to a cloud provider, which will be probably more expensive. > — [gosen on hackernews · 6 comments](https://news.ycombinator.com/item?id=50350214) > It is in the article: "the gains showed up when tokens stopped being rationed" - so it is not just a "cost per token before/after" calculation. Once you told your employees "token are free, flatrate for all - go nuts!" they started to use it for all sorts of things and increased productivity. That is not easy to calculate. > — [littlecranky67 on hackernews](https://news.ycombinator.com/item?id=50351978) **Source threads** - [hackernews](https://news.ycombinator.com/item?id=50350004) · 153 points · 151 comments ## Similar posts on daily.dev - [Leading European Financial Services Organizations Embrace Open Models to Build Proprietary Intelligence](https://daily.dev/posts/leading-european-financial-services-organizations-embrace-open-models-to-build-proprietary-intellige-ycglgxpsp) · NVIDIA · 0 upvotes · 0 comments - [I split my coding work between Claude, Qwen3-Coder and Gemma 4, and it costs less than paying for one subscription](https://daily.dev/posts/i-split-my-coding-work-between-claude-qwen3-coder-and-gemma-4-and-it-costs-less-than-paying-for-on-r1us9xfzm) · XDA Developers · 0 upvotes · 0 comments - [Building AIE CODE: The top Coding Agent Conference](https://daily.dev/posts/building-aie-code-the-top-coding-agent-conference-ovcg1zp0l) · Latent Space · 0 upvotes · 0 comments --- Tags: [#ai-coding](https://daily.dev/tags/ai-coding), [#claude-code](https://daily.dev/tags/claude-code), [#glm](https://daily.dev/tags/glm) [View this post on daily.dev](https://daily.dev/posts/revolut-moves-about-2-000-engineers-to-self-hosted-glm-for-coding-xv3uxpqtr) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Revolut moves about 2,000 engineers to self-hosted GLM for coding","url":"https://daily.dev/posts/revolut-moves-about-2-000-engineers-to-self-hosted-glm-for-coding-xv3uxpqtr","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/revolut-moves-about-2-000-engineers-to-self-hosted-glm-for-coding-xv3uxpqtr"},"datePublished":"2026-11-03T12:10:09.380Z","dateModified":"2026-11-03T20:03:38.902Z","description":"Revolut moved about 2,000 engineers onto a self-hosted GLM model for coding, making Claude Code and Codex the exception rather than the default. The setup runs...","image":"https://pbs.twimg.com/media/HWJxC2yWEAAWYYO.jpg","thumbnailUrl":"https://pbs.twimg.com/media/HWJxC2yWEAAWYYO.jpg","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/revolut-moves-about-2-000-engineers-to-self-hosted-glm-for-coding-xv3uxpqtr","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-coding,claude-code,glm","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Revolut moves about 2,000 engineers to self-hosted GLM for coding"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/revolut-moves-about-2-000-engineers-to-self-hosted-glm-for-coding-xv3uxpqtr#faq","mainEntity":[{"@type":"Question","name":"How much does Revolut pay per million tokens running a self-hosted GLM model for coding?","acceptedAnswer":{"@type":"Answer","text":"Revolut's blended cost is about 30 cents per million tokens, running on 56 Nvidia B200 GPUs. That is roughly a tenth of what the company previously paid Anthropic and OpenAI for Claude Code and Codex, which are now used as exceptions rather than the default for its roughly 2,000 engineers. Teams weighing self-hosted versus frontier AI coding tools can track real cost comparisons like this on daily.dev."}},{"@type":"Question","name":"Why did Revolut see a 70% improvement in engineering lead time after switching AI coding tools?","acceptedAnswer":{"@type":"Answer","text":"The improvement came after tokens stopped being rationed, not from the new model itself. Engineers previously saved AI usage for hard tickets, but once the self-hosted GLM setup made tokens cheap, they started using it on every ticket, which is what produced the 70% engineering lead time gain according to Revolut's AI ops lead David Abitbol. Developers rethinking when and how much to use AI assistants can follow cases like this on daily.dev."}}]} ```