GPT Image 2.5 vs GPT Image 2: release date, differences and when to upgrade

GPT Image 2.5 is the successor to GPT Image 2. OpenAI released GPT Image 2 on April 21, 2026 and GPT Image 2.5 on September 8, 2026. The request is the same: same endpoint, same fields, same size rules and the same token rates at OpenAI. What changed is that GPT Image 2.5 adds two quality steps above high, ships as two models (Flare and Sunburst) instead of one, supports transparent backgrounds outside preview, and, according to OpenAI, Flare delivers higher quality than GPT Image 2 at 50% lower latency. For a new integration, start with GPT Image 2.5 Flare; GPT Image 2 still makes sense when a tuned pipeline depends on it.

Every fact about OpenAI's models below comes from OpenAI's own pages, checked on September 29, 2026. The request details come from SeedRouter's API references for GPT Image 2 and GPT Image 2.5.

What changed from GPT Image 2 to GPT Image 2.5?

The last row comes from OpenAI's image generation guide, which now puts GPT Image 2 in a section for earlier models.

When was GPT Image 2.5 released?

On September 8, 2026. OpenAI's API changelog for that day says it "released GPT Image 2.5 Sunburst and GPT Image 2.5 Flare for image generation and editing through the Image API and the Responses API image generation tool." The same day, ChatGPT moved to ChatGPT Images 2.5 for users on every plan.

GPT Image 2 came out on April 21, 2026, alongside ChatGPT Images 2.0. Its snapshot is gpt-image-2-2026-04-21; the GPT Image 2.5 snapshots are gpt-image-2.5-flare-2026-09-08 and gpt-image-2.5-sunburst-2026-09-08.

What can GPT Image 2.5 do that GPT Image 2 cannot?

• Two higher quality steps. OpenAI's guide says the GPT Image 2.5 models "add xhigh and max quality settings" and that "earlier GPT Image models support quality settings up to high."
• A choice of speed or precision. OpenAI's Flare model page calls Flare its "fastest model for high-quality, everyday image generation." The Sunburst page says to use Sunburst "for workflows where editing precision matters most." The Flare vs Sunburst guide covers that choice in detail.
• Faster renders. OpenAI's announcement of Images 2.5 says Flare produces higher-quality images than GPT Image 2 with 50% lower latency.
• Transparent backgrounds without the preview label. For GPT Image 2, OpenAI added transparency on August 20, 2026 "in preview". GPT Image 2.5 supports it directly. On SeedRouter, set background: "transparent" with PNG output on either model.
• Steadier edits. OpenAI says GPT Image 2.5 keeps subjects from reference photos more reliably and follows edit instructions more consistently across multiple turns.

Which request parameters differ?

Almost none. On SeedRouter, both models use the same endpoint, POST /v1/images/generations, and the same fields: prompt, images, mask, size, quality, background, output_format, output_compression, n, moderation and user. Both return a task ID that you poll at GET /v1/tasks/{id}.

The differences are small:

Does GPT Image 2.5 cost more than GPT Image 2?

Not per token at OpenAI: its pricing page lists the same image and text token rates for GPT Image 2, Flare and Sunburst. The cost per image can still differ, because OpenAI notes that "token consumption can differ by model and quality setting." The new xhigh and max steps use many more output tokens: SeedRouter's reference records one 1024×1024 GPT Image 2.5 render at 196 output tokens on low and 7,024 on max.

On SeedRouter, the Standard IDs of both models charge one flat price per delivered image at any size or quality, and the Official IDs bill by tokens. For live rates and examples, see the GPT Image 2 pricing guide and the GPT Image 2.5 API pricing guide.

When is GPT Image 2 still the better choice?

• Your pipeline is tuned to it. If your prompts, sizes and review process already produce what you need at high, there is no forced reason to move. Test GPT Image 2.5 on a sample first.
• You use OpenAI's Batch API. OpenAI's model pages list Batch support for GPT Image 2, with a 50% discount, and "Not supported" for both GPT Image 2.5 models. That applies to OpenAI's own API.
• You want to compare results side by side. Because the request is identical, running the same prompt on both models is the fastest way to decide.

How do you move a request from GPT Image 2 to GPT Image 2.5?

Change model. This request works on both models once the ID is swapped:

curl https://api.seedrouter.ai/v1/images/generations \
  -H "Authorization: Bearer $SEEDROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.5-flare",
    "prompt": "An amber glass bottle on a cream background, studio lighting",
    "size": "1024x1024",
    "quality": "low"
  }'

Your existing quality values keep working, since low, medium and high exist on both. Only use xhigh or max after you switch, because GPT Image 2 does not accept them. Reference images and masks are still passed as URLs, and the finished images still come back through the same task.

Frequently asked questions

Which image model does ChatGPT use now?

ChatGPT Images 2.5, since September 8, 2026. OpenAI's help pages describe no setting to pick Flare or Sunburst inside ChatGPT, and none to switch ChatGPT back to Images 2.0.

Is GPT Image 2.5 better than GPT Image 2?

OpenAI positions it as the better model: higher quality, faster generation with Flare and more precise edits with Sunburst. For your own images, run the same prompt on both and compare.

Can I still use GPT Image 2 after GPT Image 2.5?

Yes. OpenAI still lists gpt-image-2 in its API, and SeedRouter serves both gpt-image-2 and the GPT Image 2.5 models.

Is GPT Image 2.5 free?

Only inside ChatGPT, within usage limits. The API is paid; see Is GPT Image 2 free? for what each ChatGPT plan includes.

Try both on the same prompt

Open the GPT Image 2.5 model page and the GPT Image 2 model page to run the same prompt in each playground and see the live prices. The GPT Image 2.5 API reference lists every parameter for the switch.