LLM API pricing — every model, verified and dated
Last verified
Current API pricing for 45 models across 4 providers, in dollars per million tokens. Every figure is dated and links to the provider's own documentation.
Most price tables on the web have no verification date and no source link, and LLM prices change several times a year. A number without a date is a number you cannot trust. Every model here carries the date its price was last checked, and CI fails the build when any entry goes stale.
Every current model, cheapest first
The range is 600×
Input pricing runs from $0.05 (GPT-5 nano) to $30.00 (GPT-5.5 Pro) per million tokens.
That spread is the single most important fact on this page. Choosing the right model for a task changes your bill by a multiple; almost every other optimisation changes it by a percentage. The corollary is that running everything on a frontier model because it is convenient is usually the most expensive decision in the stack.
By provider
- Anthropic — 12 models (Claude)
- OpenAI — 23 models (GPT and o-series)
- Google — 9 models (Gemini)
- xAI — 3 models (Grok)
How to read LLM pricing
- Output costs more than input, typically 3–6×, because generation is sequential and cannot be parallelised the way reading a prompt can. Estimate output volume carefully.
- Reasoning tokens are billed output you never see. A short answer from a reasoning model can carry a large output bill.
- Cached input is the biggest discount available — as low as 0.1× base input. For any workload with a stable prompt prefix, this is worth more than most model changes.
- Batch processing typically halves both rates for work nobody is waiting on.
- Token counts differ between providers. The same text tokenizes differently, so cross-provider comparisons carry a margin of error. Measure on your real inputs.
List price is not your bill
Everything here is a list rate. What you are actually charged also reflects credits, negotiated discounts, cache hits and batch usage — which is why the number that matters comes from the provider's billing API rather than a calculator.
Tracking spend across OpenAI, Anthropic, Grok and Gemini · Why billing data is daily · The OpenAI Costs endpoint