cheapinference.dev
GET STARTED

Models & pricing

Discover supported models and understand token rates.

Choose a supported model

The model catalog is generated from the same versioned price table used by the proxy. It includes 41 supported model identifiers. Search by ID or filter text and embedding models.

The GET /models endpoint is the authoritative list for the running proxy. Models that are not listed are rejected before forwarding. A listed model can still be unavailable upstream or reject model-specific parameters.

Read the pricing correctly

TermMeaning
InputUncached input tokens at the model’s standard rate.
Cached inputThe subset of input tokens served from the provider cache.
Cache writeThe subset written to cache, at the model’s applicable write rate.
OutputGenerated tokens, including reasoning tokens.
Effective cash priceList-price usage value multiplied by 0.80.

The catalog shows effective cash rates per million tokens. Credit accounting uses list-price usage value. See billing for worked examples.

Context windows and long-context rates

Model cards show the configured maximum input tokens. This is not a promise that any combination of input and output lengths will be accepted; the provider enforces each model’s constraints.

For models with long-context rates, those rates apply when the input exceeds 272,000 tokens. Standard catalog cards do not show every long-context, cached, or cache-write rate. The proxy applies the matching rate table to the returned usage.

Service tiers and unsupported modalities

Only default and auto may be requested. The actual tier returned by the provider determines the rate used in settlement. Unexpected tiers are flagged for review.

Image inputs are supported only when the selected model accepts them. This service does not provide image generation, audio, realtime, video, batch, fine-tuning, or moderation endpoints.