At AI Engineer NYC on 2026-10-13, Revolut AI ops lead David Abitbol said ~2,000 engineers now default to self-hosted GLM-5.3 on ~56 NVIDIA B200 GPUs at a blended ~$0.30 per million tokens—about one-tenth prior Anthropic/OpenAI invoices. Speaker charts claim ~60% better coding time and ~70% lead time. Claude Code/Codex remain hard-case exceptions; vision is routed elsewhere.

Key Takeaways

  • ✓Source: Revolut AI ops lead David Abitbol, AI Engineer NYC 2026 talk (2026-10-13) + official transcript; Register coverage from 2026-11-03 (speaker figures)
  • ✓Scale: ~2,000 engineers default to self-hosted GLM-5.3 on ~56 NVIDIA B200 GPUs; speaker blended cost ~$0.30 per million tokens
  • ✓Cost: invoice about one-tenth prior Anthropic/OpenAI spend; ~60% coding-time and ~70% lead-time improvements claimed (not isolated from other changes)
  • ✓Routing: GLM for text/code; vision to Gemini/GPT-class models; Claude Code/Codex kept for hard exceptions
  • ✓Ops: same GPU count tuned TTFT ~14.3s→1.7s and +76% RPS; sandbox approvals + telemetry stressed for bank compliance
⚡

Key Decision Metrics at a Glance

TL;DR Verdict
🏆 Benchmark RankRank
Cost: invoice about one-tenth prior Anthropic/OpenAI spend; ~60% coding-time and ~70% lead-time improvements claimed (not isolated from other changes)
Full Leaderboard→
💰 Cost & PlansCompare
Scale: ~2,000 engineers default to self-hosted GLM-5.3 on ~56 NVIDIA B200 GPUs; speaker blended cost ~$0.30 per million tokens
Compare Plans↗
🛠️ Ecosystem & IDERules
Claude Code (Anthropic)
Setup IDE Rules↗
🧭

Not enough VRAM? Compare cloud API and self-hosting costs

Compare 40 dev plans & simulate token costs vs $20/mo subscriptions

⚔️ Compare Model→💰 Cloud cost calculator↗
🔗

Project Links & Resources

Direct Access
Direct access to official project resources and documentation
🌐
AI Engineer NYC talk page (Revolut / Abitbol)
ai.engineer/nyc/2026/talks/revolut
↗
📖
Talk transcript (AI Engineer reading)
ai.engineer/reading/nyc-2026/transcripts/david-abitbol-revolut.md
↗
🌐
The Register: Revolut self-hosted GLM cuts coding AI cost ~10x
www.theregister.com/ai-and-ml/2026/11/03/british-fintech-revolut-slashes-ai-coding-costs-10x-by-cutting-out-claude-and-going-self-hosted/5307345
↗
🔬

In-Depth Technical Analysis

Core Background & Industry Pain Points

When teams meter Claude Code / Codex monthly, engineers save tokens for the hardest coding and skip inbox and Jira. On 2026-10-13, Revolut AI ops lead David Abitbol's AI Engineer NYC talk described making self-hosted open GLM the default so the bill is flat and rationing ends. The Register and others covered it from 2026-11-03. Figures are speaker numbers, not a lab benchmark.

Architecture Highlights & Internals

Self-hosting began around 2026-04 with GLM-5.1, then GLM-5.3 on ~56 B200 GPUs for ~2,000 engineers. Target TTFT under ~3s; ~200 tok/s is enough interactively. OpenAI-compatible APIs feed OpenCode/Codex-style clients. GLM is text-first; vision routes to Gemini or GPT-class models; Claude Code/Codex remain exceptions. Running downloaded weights on company infra differs from calling a lab-hosted API.

Authoritative Benchmarks & Measured Scores

Speaker figures: blended ~$0.30 per million tokens, about one-tenth prior Anthropic/OpenAI invoices; ~60% better coding time and ~70% better lead time (other changes not isolated). Same GPU count: ~76% higher RPS and TTFT from ~14.26s to ~1.68s. Not a third-party leaderboard—cite as Revolut talk numbers.

Developer Hands-on Guide

What transfers: tune serving to interactive latency before buying more GPUs; route vision/hard cases to frontier models; constrain agents with sandbox + approval rules and keep telemetry. Smaller teams may not hit the same unit economics. Full talk: AI Engineer; press: The Register.

Action HubReady to adopt this in production?

Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.

⚔️ Launch VS Compare→💳 View Plans↗
Up NextDigitalOcean Agent Droplets: harness + inference + 16k tools in one monthly plan from $50/$200
Read Now→