LLM API pricing

Sort the models page by blended price, or read the lower-left corner of the price map. The cheapest capable models are usually the small "flash", "mini" or "lite" tiers and open-weight models served by hosts; the cheapest route for an open model is often a host rather than the lab.

In US dollars per million tokens, with separate prices for input (the prompt) and output (what the model writes). Output usually costs 3 to 8 times input. Cached input and batch requests are cheaper where a provider offers them.

One number for ranking: three parts input to one part output, the usual mix for chat and retrieval work. Your own mix may differ; the calculator uses your actual token counts.

OpenRouter's public model catalog and its per-provider endpoint list, read every morning at 05:00 UTC, which carries each provider's list price. Each provider page links to the official pricing page. Price history before tracking began comes from Internet Archive snapshots of the same catalog.

It depends on the tier. Compare them side by side on the compare page, which shows input, output, cached and batch prices and every route (the lab, Azure, Bedrock, Vertex).

Some do. OpenRouter adds a fee when you buy credits, Vercel AI Gateway says it adds no markup, and self-hosted proxies like LiteLLM add nothing but your own hosting. The gateways page lists each one with its source.