OpenAI's own open model is cheapest on EU-sovereign infrastructure

OpenAI's gpt-oss-120b is an Apache-2.0 open-weight model — and the cheapest place to run it is not in the United States. On OVHcloud AI Endpoints (Roubaix, France; SecNumCloud-qualified) it serves at $0.09/Mtok input and $0.47/Mtok output. The identical model on US Together AI costs $0.15/$0.60 — so the European-sovereign endpoint is 22% cheaper on output and not subject to the US CLOUD Act.

Scaleway Generative APIs (Paris) hosts the same model at $0.15/$0.60, with Qwen3.6, Gemma 4 and Mistral alongside it — a full open-weight catalogue on EU-owned, GDPR- and NIS2-compliant infrastructure.

This is the heart of the repriced market: the cheapest, most portable tokens are open-weight, and the sovereign option is now also the low-cost option. The TokenFloor matrix leads with this EU Open-Weight Market; the closed US frontier sits below it as a labelled benchmark.