Frontier models.Market prices.

Available capacity competes in the marketplace. We derive your rate from capacity we can actually route to, add our margin, and cap the result at a published provider list price when one exists. If no reliable comparison price exists, we show the current rate without claiming a discount.

Catalog prices refresh when new marketplace quotes are recorded. If the live catalog cannot be reached, we label the last verified snapshot with its date instead of presenting it as live. The rate used for a request is always resolved by the API at request time.

Routing and upstream failover are best-effort, and streaming is supported. There is no SLA yet. If you need contractual uptime guarantees for critical traffic, keep a direct provider key as fallback. Your SDK's retry logic works unchanged.

Cheaper Inference records the metering and savings data needed to bill and explain each request. Optional response caching can retain request and response content privately for an account and can be disabled. The provider selected to serve a request receives its payload, just as it would when called directly. See the data-boundary section and privacy documents for the complete, current policy.

Anything OpenAI-compatible: the official OpenAI Python and JavaScript SDKs, LangChain, LlamaIndex, Vercel AI SDK, LiteLLM, or plain HTTP. Set base_url to https://api.cheaperinference.com/v1, use a ci_live_ key, and keep your existing code, streaming included.