Output tokens
The tokens a model generates in its reply. Usually the expensive half of the bill - 2 to 5 times the input price - because each one requires a full forward pass.
What it is
The model produces its answer one token at a time, each token conditioned on everything before it. This step-by-step generation is the decode phase, and it is why output is slow and costly compared with reading the prompt.
Why it matters
Output price dominates any workload that writes a lot: code generation, long-form drafting, agents that reason step by step. Output speed - tokens per second - sets how long the user waits.
Cost & infrastructure impact
Capping max_tokens and prompting for concise answers is the most direct lever on an output-heavy bill. "Think step by step" prompting improves quality but multiplies output cost.