British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hosted
British fintech giant Revolut has reportedly cut its AI coding costs by 10x by transitioning from external models like Claude and Gemini to a self-hosted GLM solution. While still utilizing GPT/Claude for vision tasks, Revolut found that the primary bottleneck wasn't code generation itself, but rather AI models handling debugging and integration with tools like Jira/Confluence. This strategic shift towards open-source and self-hosting is seen as a way to stabilize AI expenses, with other companies potentially following suit.
时间与来源
时间显示为 UTC
显示时区：UTC
本地时区尚不可用，暂时显示 UTC。
发布当时偏移：UTC+02026年11月3日 21:05 UTC
收录当时偏移：UTC+02026年11月4日 03:00 UTC
- 发布
- 2026年11月3日 21:05
- 收录
- 2026年11月4日 03:00
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 正常
档位是按信源手工设定的编辑判断，不是逐条打分。
Revolut is a fintech giant and they seem to prioritize open source and it makes their AI expenses more stable. They switched to GLM. before that, they were experimenting with Gemini and Claude. Apparently, they realized the bottleneck wasn't the code, but letting AI models deal with debugging and Jira/Confluence. They didn't eliminate GPT/Claude entirely, only using those for vision work as GLM is text only.
I can see more companies doing similar in the future, if they are already not doing it.