AI cost · OpenAI

OpenAI API cost: what you actually pay

Pure usage-based: the bill is tokens × model. The trap is running everything on the flagship, verbose prompts, and skipping caching and the Batch discount.

Public pricing

Token rates (USD per 1M tokens)

ModelInputOutputCache read
GPT-5.5Flagship; standard rates for context < 270K tokens.$5$30
GPT-5.4$2.5$15
GPT-5.4 miniLowest-cost tier for most workloads.$0.75$4.5

Batch API is ~50% cheaper; data-residency processing carries a ~10% markup; rates shown are standard processing for context under 270K tokens.

Last updated: 2026-07-22. Sourced from OpenAI API pricing (official).

Where the real OpenAI API cost hides

The sticker is the smallest part of the story. For OpenAI API, these are the line items that quietly inflate the bill:

  • Running flagship (GPT-5.5) where a mini/mid model suffices
  • No prompt caching and no Batch discount on eligible jobs
  • Runaway agent loops and untracked/leaked API keys

What an audit finds on OpenAI API

A few services pinned to the flagship model that could run on 5.4 or mini, no batching on bulk jobs, and usage with no per-team attribution.

OpenAI API pricing FAQ

How is the OpenAI API priced?

Per token, by model — e.g., GPT-5.5 at $5 in / $30 out per million tokens, down to GPT-5.4 mini at $0.75 / $4.50.

What is the cheapest way to cut OpenAI API cost?

Right-size the model, cache repeated context, and use the Batch API (~50% off) for non-interactive jobs.

What is OpenAI API really costing you?

Spendassay connects your OpenAI API spend to actual usage and engineering output — a proof-level AI cost report with the wasted dollars named. Neutral across every AI vendor.

Free · read-only · no card

More: the complete AI cost management guide · every tool, with prices · cost per outcome · compare platforms · how we measure