Output Tokens
The tokens the model generates in its response — billed at the output rate, typically several times higher than input.
Output tokens are what the model produces: the completion it streams back to you. They are billed at the output rate, which is the most expensive per-token price on almost every model — commonly three to six times the input rate, and sometimes more on reasoning-heavy models where hidden reasoning tokens are billed as output.
Output cost scales with how much the model writes, so it dominates chat, code generation, long-form drafting, and agentic loops that produce many tokens per call. A cheap input rate does not rescue a workload that generates thousands of tokens per response if the output rate is high.
When comparing models, always read the output column next to the input column — the gap between them tells you which workloads each model is priced for. This explorer shows both, plus a blended average, so you can see at a glance whether a model is tuned for reading-heavy or writing-heavy work. For reasoning models, remember that the tokens spent "thinking" count as output even when they never appear in the final answer.
FAQ
Why are output tokens more expensive?
Generating new text is more computationally expensive than reading a prompt, so providers price output higher — frequently three to six times the input rate.
Do reasoning tokens count as output?
Yes. On reasoning models, the internal reasoning tokens are billed at the output rate even though they are not shown in the final response.