GPT Image 2 is billed per token: $8.00 per million input tokens (your prompt and any reference image) and $30.00 per million output tokens (the generated image). Output cost dominates in almost every real workload, so the quality parameter and the output size are the two levers that actually move the bill. All other parameters (output_format, output_compression, background, moderation) are free.
Pricing Examples
Costs below are per call, measured from live API header deductions on April 22, 2026. Low and medium quality figures are extrapolated from the same token rates.
| Size | Quality | Typical cost per image |
|---|---|---|
| 1024x1024 | low | $0.02 |
| 1024x1024 | medium | $0.04 |
| 1024x1024 | high | $0.11 |
| 1536x1024 | high | $0.17 |
| 1024x1536 | high | $0.17 |
| 1536x1024 | auto | $0.12 |
Cost Optimization Tips
- •Use
quality=loworquality=mediumfor ideation passes. You can burn through 8 to 10 exploratory prompts for the cost of onehighpass, then upscale the winner. - •
size=1024x1024is the cheapest production size. Reach for1536x1024or1024x1536only when aspect ratio is part of the brief. - •
output_formatandoutput_compressiondo not change the price. Pickjpegat compression 90 for CDN delivery,pngat 100 when text crispness is non-negotiable. - •Edit mode (passing an
imageinput) adds input tokens for the reference. Keep reference images under 1024px on the longest edge unless you need the detail.
Example Scenario
A marketing agency shipping 20 localized Instagram Story ad variants per week at 1024x1536 and quality=high pays roughly $0.17 per image, or $3.40 per week. A production house generating 500 concert poster drafts per month at quality=medium before selecting finalists to re-render at high lands near $30 to $35 per month, all-in.