About Hubrium

What sits between your SDK and the model.

Hubrium is a routing layer in front of Chinese model providers. This page covers what it does to a request, how a rate is set, and what we publish so you can check both.

Live models
7
Model families
4
Request formats
OpenAI
Price unit
USD / 1M tokens

Request path

What happens to a request

One hop. Hubrium checks the key, resolves the model ID to an upstream provider, forwards the request, and meters the tokens the provider reports back.

  1. Your SDK

    An unmodified OpenAI or Anthropic client pointed at the Hubrium base URL.

  2. Hubrium

    Key check, model resolution, request translation and per-token metering.

  3. Provider

    The upstream model runs the request and reports its own token counts.

Transparent pricing

How a rate is set

Every price on this site is derived from the upstream cost of the model rather than a rate card we wrote. That has consequences worth stating plainly.

Rates follow upstream

When a provider changes its price, the rate here moves with it. The models page is generated from live gateway configuration, not a static table.

Unpriced models say so

A model can be reachable before it has a published rate. Those are labelled on request instead of carrying a placeholder number.

Input, cache and output are separate

Each is metered on its own. Cached input is discounted only when the provider reports an actual cache hit.

The feed is the source of truth

models.json carries the same rates this page renders, with a version hash. Build against it rather than scraping the page.

Hubrium

How we operate

  1. Publish before you ask

    Models, prices, endpoints and service status are public and machine-readable. Evaluating us should not require an account.

  2. No lock-in by design

    The API deliberately has someone else's shape. Changing one base URL is enough to leave.

  3. Say what is unknown

    Where availability or capability depends on upstream configuration, the site says so instead of smoothing it over.

Your next inference call

Read the quickstart, then decide.

The whole integration is a base URL and a key. Trying it takes less time than reading about it.