Input and output are priced differently β get your true effective rate.
Split input/output pricing means the sticker rate is not your rate β output usually dominates. Cap max_tokens to control it. Turn the per-call number into a full bill on the LLM token cost calculator.
The Blended LLM Price Calculator computes the real cost of a single API call and the effective per-1M-token rate that results from mixing input and output tokens. You enter four values: the model's input price per 1M tokens, its output price per 1M tokens, and the input and output tokens used per call. The tool multiplies each token count by its matching rate and adds the two together, so the per-call cost is driven by both the published prices and your actual token mix. Because output tokens usually cost several times more than input tokens, a call that is heavy on generated text can cost far more than its input length alone suggests.
The blended rate matters because a single sticker price hides what you will actually pay. A model with cheap input but expensive output can end up pricier than a rival once your real token ratio is applied. The practical tip: match the token counts to a representative call rather than a best case, then compare models on the blended per-1M rate instead of the headline input price. Watch the output side most closely β trimming verbose responses or capping output length often lowers cost more than shrinking the prompt.
Generating tokens is more compute-intensive than reading them, so providers price output roughly 3β5Γ higher than input. A prompt with a huge context but a short answer can still be dominated by the few expensive output tokens β or the reverse, depending on your token mix.
Your true effective rate once input and output are weighted by how many of each you actually use. Advertised prices are split; the blended rate tells you what a real call costs per token, which is what you should use for margin and forecasting.