Compare models side by side
Up to 5 models: what each would cost for your workload, and how their prices, limits and features differ.
What are you building?
Your workload
How much do you send each time?
How long is each answer?
Does every request start with the same text?
The same instructions every time, like a system prompt.
Providers discount repeated text (prompt caching) only when the repeated part is long enough, from 512 to 4,096 tokens depending on the model, and was sent recently.
I know my exact token counts
Same text % is the share of input tokens that is identical on every request.
Add models above, or start from one of these matchups: