Revolut details its AI coding stack: per-token billing plus 4-bit weights and 90%+ cache hits
ccerrato147 · x · 2026-11-05
Fintech firm Revolut is one of the few companies still paying for Codex, Claude Code and Gemini. David Abitbol laid out the strategy at AI Engineer (Oct 29): pay nothing for the default, pay per token only for the exceptions.
The quoted post details the engineering: none of it came out of the box. On the same 56 GPUs, time-to-first-token dropped from over 10 seconds in April to 1.6 by August while serving more than twice the traffic. Tactics included capping context at 256K tokens, running weights at 4 bits, and tuning for cache hit rates above 90%.
More from coding & agent
- "Evals > vibes": agent workflow debates are just guessing without eval stats — annetgriffin · 2026-11-05
- OpenAI hackathon winner joins YC, launches Raven, an agent-first phone — ycombinator · 2026-11-05
- Open-source Cicada brings real-time translation and AI meeting notes to any Mac call, free 50h/month — dotey · 2026-11-05
- IBM launches AI coding assistant named "Bob", devs can't stop laughing — zeeg · 2026-11-05
- Zach Lloyd: Don't vibe code your software factory — prompt-chained agents don't scale — vikvang1 · 2026-11-05
- Grok bot phone call measured: 85-second real call costs just 0.2% of quota — elonmusk · 2026-11-05