Is DALL-E 3 Still Available? What Replaced It in 2026

Is DALL-E 3 Still Available? What Replaced It in 2026
Quick answer

DALL-E 3 was removed from OpenAI's API on May 12, 2026, and is no longer accessible for new projects. It was replaced by GPT Image 2 (branded "ChatGPT Images 2.0" inside the ChatGPT app), which offers better prompt accuracy, higher resolution, and more editing control than DALL-E 3 ever had.

Millions of people still search for DALL-E 3 — but it's gone. Here's exactly when it was retired, what replaced it, and where to generate images with OpenAI today.

Is DALL-E 3 Still Available? What Replaced It in 2026

Is DALL-E 3 still available in 2026? No. OpenAI retired it on May 12, 2026, both in the API and in ChatGPT, yet millions of people are still searching for it, comparing it, and trying to work out what to use instead. This guide gives the direct answer, the exact timeline, what the replacement costs, how to move existing code over, and which other image generators are worth considering now.

The Full Timeline

November 14, 2025: Deprecation announced

OpenAI gave developers roughly six months' notice that the DALL-E 3 API endpoint would be shut down.

December 2025: ChatGPT quietly switched models

Most users never noticed, but OpenAI swapped DALL-E 3 for an interim model, GPT Image 1.5, behind the scenes inside ChatGPT.

April 21, 2026: GPT Image 2 launched

Marketed to consumers as "ChatGPT Images 2.0," this became OpenAI's new flagship image model. It added a "Thinking Mode" that can search the web mid-generation and produce up to 8 coherent images from a single prompt.

May 12, 2026: DALL-E 3 fully retired

The API endpoint was shut down entirely. Any code still calling the old dall-e-3 model ID now returns an error. DALL-E 2 was retired on the same date.

What Actually Replaced DALL-E 3

DALL-E 3 (retired) GPT Image 2 (current)
Status Retired May 12, 2026 Active flagship model
Resolution Standard, capped near 1,024 px Native 2K, 4K in beta
Prompt accuracy Good Significantly improved
Text rendering Decent Industry-leading
Multi-image output One image at a time Up to 8 coherent images per prompt
Access Was ChatGPT Plus + API ChatGPT Plus, Pro, Business + API
Pricing (API) ~$0.04/image (flat) $0.005–$0.211/image (token-based, varies by quality/resolution)

The short version: GPT Image 2 is a different architecture, not a patched DALL-E 3. OpenAI built it for stronger prompt following, accurate in-image text, and multi-image consistency, which is why it chose a full replacement instead of an incremental update to the older model.

How to Access It Today

Through ChatGPT

If you already have ChatGPT Plus ($20/month) or Pro, image generation now runs on GPT Image 2 automatically, so there is nothing extra to enable. Free tier users get a more limited "Instant Mode," while the advanced "Thinking Mode" is reserved for paid plans.

Through the API

Developers migrating existing DALL-E 3 integrations need to switch to one of the current model IDs: GPT Image 2, GPT Image 1.5, or GPT Image 1 Mini for lower-cost, high-volume use cases. Older GPT Image variants are also reported to be on a retirement schedule later in 2026, so new builds are safer targeting GPT Image 2 directly. Check OpenAI's deprecations page before committing to a date.

What GPT Image 2 Costs Through the API

DALL-E 3 charged a flat fee per image. GPT Image 2 bills by token instead, at roughly $30 per million image output tokens and $8 per million image input tokens. In practice, a standard 1024×1024 image lands at about $0.006 on low quality, about $0.053 on medium, and about $0.211 on high. Portrait and landscape sizes come in a little lower, around $0.005, $0.041, and $0.165.

Two things catch people out. First, a single prompt can cost forty times more on high quality than on low, so set the quality parameter deliberately instead of leaving it at the default. Second, editing with reference images costs more than plain text-to-image, because the input images are billed as tokens too. Heavy edit workflows can run two to three times the baseline estimate. Always check OpenAI's live cost calculator before budgeting a large batch.

Migrating Code From DALL-E 3: A Practical Checklist

  • Change the model ID. Replace dall-e-3 with the GPT Image model you picked. The old ID returns an error.
  • Re-test your size and quality settings. The parameters differ from the flat DALL-E 3 options, and quality now drives price directly.
  • Check how images come back. GPT Image models return image data differently from DALL-E 3's hosted URLs, so code that downloaded from a URL may need changing. Confirm against the current API reference.
  • Re-test your prompts. GPT Image 2 follows instructions more literally, so prompts tuned around DALL-E 3's quirks can produce different results.
  • Add a cost cap. Token billing means a runaway loop gets expensive faster than it did with a flat per-image fee.

Should You Even Use OpenAI's Image Model?

GPT Image 2 is a real improvement over DALL-E 3, but it is not automatically the best option for every job. Here is how the main alternatives compare:

Tool Best for Free tier
GPT Image 2 Accurate in-image text, easy access inside ChatGPT Limited, via ChatGPT free
Midjourney Artistic composition and stylized output No
Nano Banana Pro Photorealism and text accuracy, closest rival to GPT Image 2 Yes, through the Gemini app
Ideogram Posters, logos, and anything with lettering Yes, limited daily prompts
Adobe Firefly Designers already working in Adobe apps Limited monthly credits
Stable Diffusion Full control and no usage cap if you run it yourself Free, needs your own hardware

Free-tier limits change often, so treat the last column as a starting point and confirm on each vendor's pricing page. For a deeper look at two of these, read our Nano Banana Pro review and our Ideogram review.

Which One Should You Pick?

If you already pay for ChatGPT Plus and just need reliable images with correct text, stay with GPT Image 2: there is no extra tool and no extra subscription. If photorealism matters most and you want a free option, try Nano Banana Pro first. If you want artistic, stylized work and are happy to pay, Midjourney is still generally considered the stronger pick. If your images are mostly text-heavy graphics, Ideogram is built for that job.

Ready to compare more options? Browse every image generator in our AI tools directory, or read more guides in the blog.

Sources

Checked on November 15, 2026. OpenAI's own API pricing page blocks automated access, so check live image prices there:

Frequently Asked Questions

Can I still use DALL-E 3 through the API?

No. Any request using the dall-e-3 model ID returns an error as of May 12, 2026. You need to migrate to GPT Image 2, GPT Image 1.5, or GPT Image 1 Mini.

Is GPT Image 2 free?

It's included with ChatGPT Plus ($20/month), Pro, and Business plans. Free ChatGPT users get a more limited version through Instant Mode, without the advanced Thinking Mode features.

Why did OpenAI retire DALL-E 3 instead of just updating it?

OpenAI moved to a new model architecture built for stronger prompt following, better text rendering, and multi-image generation — improvements significant enough that a full model replacement made more sense than an incremental update to the older architecture.

What's the closest free alternative to DALL-E 3?

Nano Banana Pro (Gemini 3 Pro Image) has a genuine free tier through the Gemini app, roughly 2 images a day, and is currently the closest rival to GPT Image 2 on photorealism and text accuracy. For stylized or artistic output instead of realism, Midjourney is the stronger pick, though it doesn't offer a free tier.

How much does GPT Image 2 cost per image through the API?

It is billed by token, not a flat fee. At 1024×1024 an image costs roughly $0.006 on low quality, $0.053 on medium, and $0.211 on high. Edits with reference images cost more because the input images are billed too, so check OpenAI's cost calculator before a large batch.

Do I need to change my code now that DALL-E 3 is retired?

Yes, if it calls the dall-e-3 model ID. Swap in a current GPT Image model ID, re-test your size and quality settings, and confirm how your code reads the returned image, since GPT Image models handle the response differently from DALL-E 3's hosted URLs.

Share this article:

User Reviews

Be the first to review!

Leave a review