GPT-5 API cost calculator

How GPT-5 API pricing works

GPT-5 is billed per token, split into input (your prompt plus any context you send) and output (everything the model writes back). On each tier, output costs several times more than input — GPT-5 is $1.25 / 1M input and $10.00 / 1M output, an 8× gap — so the length of the answer is the single biggest lever on your bill. Capping max_tokens, asking for concise responses, and trimming system prompts all cut cost directly. Because GPT-5 is a reasoning model, the hidden reasoning tokens it generates before the final answer are billed as output too, so heavy step-by-step reasoning shows up on the bill even when the visible answer is short.

The second lever is tier choice. The family spans a wide price range: flagship GPT-5.5 at $5.00 / $30.00 for the hardest reasoning and long-form work, GPT-5 as the balanced default, GPT-5 mini at roughly a third of GPT-5's price for everyday tasks, and GPT-5 nano at $0.10 / $0.40 for classification, extraction and high-volume short replies. A common pattern is to run a cheaper tier by default and reserve GPT-5 or GPT-5.5 only for the requests that genuinely need frontier quality — the table above shows exactly what each tier costs at your volume. To see GPT-5 against Claude, Gemini and the rest, use the LLM price comparison; for a from-scratch breakdown of how per-token billing adds up, read how LLM API pricing works or browse the Learn hub.

FAQ

What's GPT-5's cost per 1M tokens? $1.25 input and $10.00 output (reference, 2026). GPT-5 mini is $0.40 / $1.60 and GPT-5 nano is $0.10 / $0.40.

Which GPT-5 tier should I use? Match the tier to the task — nano or mini for routine, high-volume calls, GPT-5 for balanced work, GPT-5.5 only for the hardest requests.

How do I lower a GPT-5 bill? Shorten outputs, cap reasoning, cache repeated context, and route easy calls to GPT-5 mini or nano.