Models

# Models

Every frontier model sits behind one OpenAI-compatible surface. You choose a model by name; Graphene handles provisioning, routing and billing. Switching models is a string change, not an integration.

## Naming conventions

Model identifiers follow the provider's own naming so your existing code ports unchanged — for example `claude-opus-4-8`, `gpt-...`, or `gemini-...`. Pass the identifier as the `model` field exactly as you would against the provider's own API. The full, current catalogue is on the [models page](/models).

## How pricing flows through

Graphene bills per token at a flat **20% under each provider's published list price**. When a provider changes its list price, the routed rate you pay moves with it automatically — there is no separate Graphene price sheet to reconcile. The rate applied to every request is shown alongside its token counts in **Usage**, so the discount is verifiable per call rather than promised in the abstract.

## Portability is the default

Because the surface is OpenAI-compatible and models are addressed by their provider names, moving on or off Graphene is the same one-line base-URL change described in the [Quickstart](/docs/welcome). There is no lock-in to design around.