Documentation

Merius API

Merius serves open models on our own Blackwell GPUs over one OpenAI-compatible API. If you have written against the OpenAI SDK, you already know the request shape — change the base URL, use an Merius key, and call any model we serve.

Introduction #

The Merius API speaks the OpenAI HTTP schema end to end: the same endpoints, the same request and response bodies, the same streaming format. You do not install an Merius SDK or learn a new client — any OpenAI-compatible library works once you point it at our base URL and pass an Merius API key.

Under that familiar surface, every request runs on hardware we own and operate: B200/B300 (NVIDIA Blackwell) GPUs in EU and US datacenters. You get transparent per-token pricing, EU data residency on request, and no prompt or completion retention.

Already using the OpenAI SDK? Skip ahead to Authentication and Quickstart — the only lines that change are your base URL and key.

What you get #

A short list of what the API gives you, and where to read more:

OpenAI-compatible

Drop-in /chat/completions, /completions, and /models endpoints. Keep your SDK and request shape; change one line.

Open models, our GPUs

Qwen3, GLM, and Mistral open models served on owned B200/A100 hardware — not resold cloud capacity. See Models.

Streaming, tools, JSON

Server-sent-event streaming, function calling, and structured outputs, all in the OpenAI format you already use.

EU + US, zero retention

EU-only routing on request, GDPR-compliant, no prompt or completion storage. See Data residency.

Make your first request #

Three steps and you are calling a model:

A complete request
curl https://api.merius.ai/v1/chat/completions \
  -H "Authorization: Bearer $MERIUS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen3-30b-a3b",
    "messages": [{"role": "user", "content": "Hello"}]
  }'
  • Create an Merius account and generate an API key in the dashboard.
  • Point your OpenAI client at the Merius base URL (the one line that changes).
  • Call /chat/completions with any model slug we serve.

The Quickstart shows the same call in Python and TypeScript. Authentication covers creating and storing keys.