Skip to content
Multigrid
Providers

List your models on Multigrid

If you run inference and want another channel, here is what integrating involves, what it gets you, and where we honestly are today.

Models in the catalogue
304

Live count, not a claim

Enabled routes
338

Every one price-checked

Providers integrated
6

Each with measured uptime

Where we actually are
Multigrid has not launched, so there is no meaningful traffic to route to you yet. There is no provider agreement drafted, and we have never paid a provider by invoice against metered usage — every provider above is one we hold a prepaid account with. For most providers the sensible answer today is “interesting, come back with volume”, and we would rather you reach it here than three emails in.

How it goes

  1. 1

    Tell us what you serve

    Models, prices, and whether the API is OpenAI-compatible. One message; there is no form to fill in and no qualification call.

  2. 2

    We write the route

    An adapter if your wire format needs one, otherwise a catalogue entry. Prices are recorded as costs and your uptime starts being measured from the first request.

  3. 3

    You appear in the catalogue

    On /providers and on every model page you serve, with your real prices and your measured uptime, ranked against the others by price and health rather than by any deal.

What we need from you

Four things, taken from the adapter contract the gateway actually runs.

An OpenAI-compatible API

Or Anthropic's or Google's shape — those are the three adapters. If you already speak /chat/completions, adding you is a catalogue row rather than new code.

Streaming with usage in it

SSE, and a usage object carrying prompt and completion counts. We bill from what the response reports, never an estimate, so a stream that never states its usage cannot be priced honestly.

Per-token prices

Prompt and completion, per million, and a way to notice when they change. They are published as costs on your model pages rather than hidden behind a number of ours.

A models endpoint

Anything we can seed a catalogue from — context length, modality, max output. An endpoint is easiest because it doubles as a credential check.

How your prices are treated

Published as costNo spread

Every route on /models shows what you charge. Our margin comes from fees on account top-ups, not a spread on your tokens, so a customer comparing you against another provider here is comparing your real number.

Routing is by price, health and capability. Nothing about position is for sale — which cuts both ways, since nothing hides a bad week either. Uptime is measured and published per provider on /providers.

What a route looks like

one catalogue entry
{
  "provider": "your-slug",
  "model": "your-org/your-model",
  "adapter": "openai",
  "base_url": "https://api.example.com/v1",
  "prompt_price_micros": 150,
  "completion_price_micros": 600,
  "context_length": 128000
}

Start a conversation

Use the contact form and start the subject with Provider. Lead with anything that would have to be true before you would sign — a minimum volume, a region, an exclusivity you cannot give. It may end the conversation early, which is better for both of us. One person reads these.

Get in touch