Artificial intelligence token pricing commonly assigns rates per million input and output tokens. Providers may also distinguish cached input, long context, batch processing or particular model tiers.
A low token price can reduce raw inference cost, but it does not guarantee the lowest cost for a business outcome. A weaker model may need longer prompts, more output, repeated attempts or human correction to complete the same task.
Acronyms and aliases
AI token pricing acronymprice per token synonymartificial intelligence token pricing variant
Related terms
Frequently asked questions
Why are output tokens often more expensive than input tokens?
Autoregressive generation processes output sequentially and can consume more serving time per token than parallel prompt processing.
Should a business choose a model only by token price?
No. It should compare the complete cost, quality, reliability and number of attempts required to produce an acceptable result.
