Breaking

Revolut Says Self-Hosted Chinese Model Cut Its AI Coding Costs About Tenfold

November 4, 2026 at 20:50 EST

Software Dev & Coding
Finance
Sales & Business

Revolut says it has cut its AI coding costs by roughly 10x by making a self-hosted Chinese model its default in place of Anthropic's Claude Code.

Revolut · AI coding costs

Revolut claims a roughly 10× cut in AI coding costs by switching to a self-hosted Chinese model

The fintech reportedly made Z.ai's GLM its default coding assistant in place of Anthropic's Claude Code, running the model on its own Nvidia GPUs at about $0.30 per million tokens.

~10×

Claimed reduction in AI coding costs

$0.30

Approximate cost per million tokens on self-hosted GLM

Independent benchmarks comparing GLM with Claude Code identified

Relative coding cost, scaled to the claim

Each block is one unit of cost. The prior setup is shown at 10 units, the self-hosted setup at 1, matching the claimed ~10× difference.

10×

Claude Code (prior default)

1×

Self-hosted GLM (claimed)

Where the spending moves

Per-token fees

Usage bills grow with every request

→

Fixed infrastructure

Owned Nvidia GPUs running the model

→

Staffing and utilization

Savings depend on how busy the hardware is and the engineering effort to run it

A large claim that remains unverified

The cost figure is the company's reported number. The points below are not yet established.

Output quality

Whether GLM matches Claude Code is not established.

Full cost

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.