Token-based pricing

Each API request is metered: the tokens in your prompt, including the system prompt, conversation history and any documents, are billed at the input rate, and the tokens the model generates are billed at the output rate, which is usually higher. Providers may also offer lower rates for cached input that repeats across requests and for batch jobs that can wait, and reasoning models may bill their internal reasoning as output tokens.

Why it matters: two applications using the same model can have very different bills depending on prompt length, how much history they resend and how long the answers are. For buyers of AI-powered software, token costs often sit behind the credits, usage caps or fair-use limits of a subscription. Understanding the meter helps you forecast spend and compare offers.

Example: a support bot handling 10,000 conversations a month, each with about 2,000 input and 500 output tokens, uses 20 million input and 5 million output tokens; multiplying those by the provider's published rates gives the monthly model cost.

Limits: rates change frequently and vary by model, region and tier, so always use the provider's current pricing page. Token counts also differ between tokenizers, which makes cross-provider comparisons approximate.