← Back to articles

Image generation API cost guide

Image generation API pricing: compare the bill, not the headline rate

Some providers charge for generated images. Others bill separate text-input, reference-image-input, and generated-image tokens. A rate per million tokens is not a price for one image, and a per-image rate is not a token rate.

Published Nov 8, 2026Official sources checked Nov 8, 2026

Billing meters

Two formulas, different denominators

Per generated image

Output fee = billable images × rate per image. Check whether size, region, input images, or successfully returned outputs change the bill.

Output image tokens

Output fee = billed image-output tokens ÷ 1,000,000 × image-output token rate. Quality and dimensions may change token use.

Inputs and edits

Add billed prompt text tokens and reference-image tokens, if applicable. Do not apply a cache or Batch discount to an endpoint without confirming eligibility.

Compare totals for the same workload and currency. A displayed rate alone does not establish cost per successful asset.

Token-priced example

Estimate one GPT Image 2.6 request

OpenAI publishes Standard rates for GPT Image 2.6 of $5 per million text-input tokens and $30 per million image-output tokens. Image inputs, when provided, have their own $8 per million image-input tokens meter.

Illustrative output

3,000 image-output tokens × $30 / 1M = $0.090

Illustrative prompt

200 text-input tokens × $5 / 1M = $0.001

Illustrative total

$0.091 for this assumed token usage, with no reference-image input

The 3,000 and 200 token counts are hypothetical, not a fixed token count for a particular image size or quality. Use actual usage or OpenAI's image-cost estimator for the chosen settings. OpenAI documents Responses-tool-only cached-input eligibility for GPT Image 2 and 2.5. Do not assume the same discount applies to direct Images API requests or to GPT Image 2.6 without a route-specific official confirmation.

Official sources: OpenAI image generation guide and estimator, OpenAI pricing.

Per-image example

Keep Qwen Image's regional price in CNY

Alibaba Cloud lists qwen-image-2.1-pro image generation by generated image, not by OpenAI-style image-output tokens. Its model page lists CNY 0.25 per image in Japan (Tokyo) and CNY 0.283404 per image in Singapore, before any eligible promotion.

Tokyo: 100 outputs

100 × CNY 0.25 = CNY 25.00

Singapore: 100 outputs

100 × CNY 0.283404 = CNY 28.3404

Compare like for like

Choose the same model route, region, output count, resolution, and billing conditions first. Convert currency only with a stated exchange rate and date.

Alibaba's general image billing policy charges for successfully generated outputs; consult each model's input-image terms separately. Do not assume that all provider failures and retries follow the same policy.

Official sources: Qwen Image 2.1 Pro model card, Alibaba Cloud image billing rules.

Production budgeting

Five inputs to record for every image quote

  1. Exact route: provider, model ID, endpoint, deployment region, and currency.
  2. Output target: image count, resolution, quality, and number of returned variants.
  3. Input workload: prompt length, reference images, edit passes, and billed input units.
  4. Delivery behavior: successful outputs, chargeable retries, streaming/partial-image fees, and applicable discounts.
  5. Evidence: official pricing URL, read date, and a measured usage or invoice sample before scaling.

Use the image-generation calculator for verified per-image or image-token estimates. Some routes lack the pricing data needed for an estimate; unknown does not mean free. Browse the image-generation catalog for model routes and sources.

Source and update policy

Official OpenAI and Alibaba Cloud documentation was checked November 8, 2026. Prices, regions, endpoint eligibility, and output token counts can change. Recheck primary sources for your exact route before committing a production budget.