World Model Optimizer launches a router that cuts agent inference cost by 40%+

SilenN · hn · 2026-07-27

World Model Optimizer launched wmo serve, a tool that routes repetitive tasks to smaller distilled models while preserving frontier-like quality at roughly half the cost.

What it does

The system takes agent traces plus an OpenRouter key, spins up a local OpenAI-compatible endpoint, and decides which requests should go to the frontier model versus the cheaper specialized one. The company also says a hosted version is available at 40%+ lower cost, with self-improvement over time.

Original post →

More from Infra

Infra channel →