GPT-Image-2 API is now an official OpenAI developer topic. OpenAI lists gpt-image-2 as the model ID, names gpt-image-2-2026-04-21 as the current snapshot, and uses the model in Image API generation, edit, streaming, and mask examples.
The release question is therefore no longer whether a public model row exists. The useful question is what should change in your code, cost model, provider routing, and older GPT-Image-2 articles now that the official contract is visible.
This page was rewritten after OpenAI published the GPT Image 2 model page and pricing evidence. Older release-watch wording is no longer the live contract.
Quick Answer
| Question | Answer |
|---|---|
| Is GPT-Image-2 API released? | Yes. OpenAI now documents gpt-image-2 and snapshot gpt-image-2-2026-04-21. |
| What should code call? | Use model: "gpt-image-2" on Image API generation or edit calls when your organization is eligible. |
| Which route for assistant flows? | Use Responses API with a mainline model and the image_generation tool; do not put gpt-image-2 in the Responses model field. |
| What changed from old pages? | The old release-watch boundary is obsolete; official model, examples, and pricing now exist. |
Official Evidence
| Evidence | Current contract | Source |
|---|---|---|
| Model ID | gpt-image-2 | OpenAI model page |
| Snapshot | gpt-image-2-2026-04-21 | OpenAI model page |
| Direct API | /v1/images/generations and /v1/images/edits examples use model: "gpt-image-2" | Image generation guide |
| Assistant flow | Use a mainline model such as gpt-5.4 with the image_generation tool; gpt-image-2 is the GPT Image model behind the image process, not the Responses model field. | Responses image tool |
| Pricing | Standard rows: image input $8, cached image input $2, image output $30; text input $5, cached text input $1.25, text output $10 per 1M tokens. | Pricing |
| Cost estimate | 1024x1024: low $0.006, medium $0.053, high $0.211; portrait/landscape examples are $0.005, $0.041, $0.165. | Cost calculator |
These facts are enough to change the live answer from release watch to released. Release still does not mean every account can call every mode without eligibility, organization verification, or current rate-limit checks.
Release Date Answer: Public, but Still Account-Gated
As of April 22, 2026, the public evidence is the gpt-image-2-2026-04-21 snapshot plus OpenAI developer docs showing the model page, Image API examples, and pricing rows. The title can answer released, but the body still needs an account boundary: organization verification, model access, rate limits, and regional or compliance settings may affect actual calls.
| Question | Answer |
|---|---|
| What date should I cite? | Use the snapshot date 2026-04-21 and the page verification date 2026-04-22. |
| What changed in code? | Image API calls can use model: "gpt-image-2"; Responses workflows should use a mainline model with image_generation. |
| What should old articles stop saying? | They should stop saying no public API row, no release date, or GPT Image 1.5 is the only current public anchor. |
From Release Watch to Migration Check
The valuable part of the old page was not the conclusion; it was the proof standard: model page, endpoint example, pricing row, and official wording. Now that those signals exist, the reader job becomes migration. Confirm the route first: Image API for direct generation and edits, Responses for multi-turn assistant flows. Then confirm output format, size, quality, cost logging, and retry behavior before shipping a model-string change.
How SEO and GEO Pages Should Change
Search queries will still contain release date, when released, and GPT-Image-2 API wording, but the first screen should not echo stale search language. Title, description, excerpt, FAQ, and answer blocks should move the reader to the current fact: public model, model ID gpt-image-2, snapshot gpt-image-2-2026-04-21, and available pricing and examples. GEO needs a dated, quotable answer, not a rumor timeline.
Short Answer for AI Citation
GPT-Image-2 API is public in OpenAI developer docs as gpt-image-2, with snapshot gpt-image-2-2026-04-21. Use Image API for direct generation or edits, and Responses API with image_generation for assistant workflows.
Related next step
FAQ
When was GPT-Image-2 API released?
Use the public snapshot gpt-image-2-2026-04-21 and the verification date April 22, 2026. OpenAI docs now show model, examples, and pricing.
Can I call gpt-image-2 now?
Yes on the Image API when your organization and account are eligible. Test your own key before changing production traffic.
Does this replace GPT Image 1.5?
For pages about the newest official model, yes. For existing systems, treat migration as a tested model/config update.