Revolut Reports 90% Cut in AI Coding Costs After Switching to a Self-Hosted Open-Weight Model

·2 views

Revolut, the British fintech, rolled out AI coding assistants to about 2,000 engineers, but spending on tokens grew costly, with some individuals running up large monthly bills. In April 2026 the company began self-hosting open-weight models, starting with GLM-5.1 and later GLM-5.3, served on around 56 Nvidia B200 GPUs. According to figures David Abitbol presented at the AI Engineer conference in New York, this reduced inference costs by roughly 90% and improved coding speed by about 60%. Frontier tools such as Claude Code and OpenAI's Codex remain available as a fallback for tasks the self-hosted model cannot handle, such as text-only limitations on UI work. The figures are the company's own, and they exclude cloud, staffing, and observability expenses, while the infrastructure required is substantial.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Python List Methods Explained: Adding, Removing, Searching and Sorting

A tutorial on DEV Community, originally published on CodingCush, walks through Python's built-in list methods. These functions let programmers add, remove, search, order and copy items without writing lengthy code. The guide covers append, extend and insert for adding items, remove, pop and clear for deleting them, and index and count for searching. It also explains that most methods modify the original list in place and return None, a common pitfall for beginners. Each method is illustrated with short, runnable examples and cautions about errors such as ValueError and IndexError.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Guide Shows How to Prepare for a Junior Backend Interview Using a Single Project

A developer guide argues that candidates preparing for junior backend roles can use one personal project, rather than chasing every technology in job listings, to cover multiple interview topics. It suggests a simple project such as a task management API built with Node.js, Express, and PostgreSQL, which already touches on REST endpoints, data handling, and authentication. Candidates are encouraged to walk through request flows, justify database design choices, and explain how login and protected routes work. The article also recommends deliberately testing edge cases like missing fields, invalid input, and unavailable databases, then rehearsing answers to a prepared question list without consulting the source code.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Founder Launches VersaShield AI to Help Users Spot Online Scams and Phishing

Deeya, a young founder, has introduced VersaShield AI, a tool designed to make digital safety more accessible. The product analyzes suspicious messages and links and helps users check potential scams and fake payment claims. It also includes QR code and image analysis to flag risky content. The developer emphasizes privacy-first design with local processing and says the aim is to give everyday users simple protection without requiring technical knowledge. The project is presented as ongoing, with the founder seeking feedback from the developer community.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Ghostlet: A Pixel-Art Desktop Companion Built for Hacktoberfest 2026

Three friends created Ghostlet, a small pixel-art desktop app for Windows, as a Hacktoberfest 2026 Week 4 project. When a user's computer sits idle, ghosts appear around the cursor, accompanied by pumpkins, graves, and other spooky decorations that build up over time. Any mouse movement or keypress causes the ghosts to scramble away with their decorations. The project was written in Python using PySide6 and the Windows API, with AI assistance used as a learning aid, and it is open source under the MIT License.

0 comments Read more at DEV Community