DALL-E 3 Review 2026: What It Did Well, What It Cost and Why OpenAI Retired It

DALL-E 3 left the OpenAI API on May 12, 2026 and ChatGPT on August 30, 2026. Our review covers its strengths, its old per-image prices and what to use instead.

The Proof Desk9 min read

DALL-E 3 launch illustration of a red-cloaked woman haggling with a top-hatted vendor on a moonlit city street, with callouts linking each detail to the prompt

If you typed openai.com/dall-e-3 into a browser in 2026, you were looking for a product OpenAI no longer sells. DALL-E 3 was retired on Azure OpenAI on March 4, 2026, removed from the OpenAI API on May 12, 2026, and the official DALL-E GPT inside ChatGPT closed on August 30, 2026.

That does not make it irrelevant. DALL-E 3 was the model that put image generation inside ChatGPT for millions of people, and it set the template its successors still follow: describe what you want in plain language, let the chat model write the real prompt, and get back an image that matches the request. This review covers what DALL-E 3 did well, what it cost, why it was replaced and what to use now.

Every date and price below comes from OpenAI, Microsoft or Bing documentation, with archived copies where the live page has changed. Where only third-party sources exist, we say so.

Where DALL-E 3 stands in October 2026

The short answer: you cannot buy DALL-E 3 from OpenAI any more, and the main ways to reach it have closed one after another.

Azure OpenAI: Microsoft's documentation says dall-e-3 "was retired on March 4, 2026, and is no longer available for new deployments. Existing deployments are non-functional."
OpenAI API: the dall-e-2 and dall-e-3 snapshots were "deprecated and removed from the API on May 12, 2026", after developers were notified on November 14, 2025.
ChatGPT: OpenAI retired the official DALL-E GPT on August 30, 2026 and pointed users to ChatGPT Images. User-created GPTs with image generation enabled were not affected.
Bing Image Creator: Microsoft added GPT-4o next to DALL-E 3 in August 2025 and let users pick either model. Several 2026 guides report DALL-E 3 is still listed there, but Microsoft has not confirmed its current status, so treat it as a possibility, not a plan.

If you still have DALL-E 3 code

Requests to dall-e-3 no longer work. OpenAI's current dall-e-3 model page recommends GPT Image 2.5 Sunburst. The deprecation table also names gpt-image-1 and gpt-image-1-mini, but gpt-image-1 retires on October 23, 2026 and gpt-image-1-mini on December 1, 2026, so skip them and migrate once.

What DALL-E 3 was and how it worked

OpenAI announced DALL-E 3 in September 2023. Its launch page promised it to ChatGPT Plus and Enterprise customers in early October, and the ChatGPT release notes for October 16, 2023 describe it rolling out in beta, chosen from a selector under GPT-4. API access followed at DevDay on November 6, 2023.

The key design choice was that DALL-E 3 was "built natively on ChatGPT". You did not have to learn prompt syntax. In OpenAI's words, "ChatGPT will automatically generate tailored, detailed prompts for DALL-E 3 that bring your idea to life." If the first result missed, you asked for a change in a follow-up message and ChatGPT rewrote the prompt for you.

ChatGPT conversation asking what a super-duper sunflower hedgehog looks like, answered with four DALL-E 3 hedgehog images — DALL-E 3 ran inside ChatGPT, which turned a plain request into a detailed prompt. Screenshot from OpenAI's 2023 launch page.

Access widened over time. On August 8, 2024 OpenAI let ChatGPT Free users create up to two images per day with DALL-E 3, one of the few hard numbers OpenAI has ever published for free image generation.

In the API it was a plain text-to-image model: text in, image out, at three fixed sizes (1024x1024, 1024x1792 and 1792x1024) in standard or HD quality.

Image quality: what DALL-E 3 did well

Against DALL-E 2 the jump was large. The two biggest gains were prompt adherence and detail: DALL-E 3 followed long, specific descriptions instead of latching onto a few keywords, and it produced cleaner textures, lighting and composition.

Side by side oil paintings of a basketball player dunking as a nebula explosion, DALL-E 2 on the left and DALL-E 3 on the right — Same prompt, two models: "An expressive oil painting of a basketball player dunking, depicted as an explosion of a nebula."

It could also place short, legible text inside an image, something DALL-E 2 rarely managed, though longer passages still came out garbled.

Art Deco travel poster in gold and black reading Explore Venus, generated by DALL-E 3 — The headline lettering came out clean, a big step up from DALL-E 2, but the small print at the bottom is still gibberish. Sample from OpenAI's launch gallery.

Where it fell short

Fixed sizes. Only three output sizes were available; flexible image sizes arrived with GPT Image 2.
Text in, image out. DALL-E 3 accepted only text input, so editing from reference photos and high-fidelity image inputs had to wait for its successors.
Prompt rewriting cut both ways. ChatGPT's expanded prompts helped beginners, but the prompt the model actually received was not exactly what you wrote.
No C2PA labels at launch. The launch page described an internal provenance classifier; C2PA metadata on every image arrived with GPT-4o image generation in March 2025.

Safety limits and who owns the images

DALL-E 3 shipped with stricter refusals than its predecessor. OpenAI said it had "mitigations to decline requests that ask for a public figure by name" and was "designed to decline requests that ask for an image in the style of a living artist." Creators could also opt their images out of training for OpenAI's future image models.

On ownership, the launch page was unusually direct: "the images you create with DALL-E 3 are yours to use and you don't need our permission to reprint, sell or merchandise them." OpenAI's Terms of Use effective January 1, 2026 still name DALL-E: you own the output, and OpenAI assigns you "all our right, title, and interest, if any" in it. The same terms warn that output "may not be unique", so another user may get something similar.

DALL-E 3 cartoon of a hollow avocado on a therapist's couch saying I just feel so empty inside, with a spoon-headed therapist taking notes — One of OpenAI's own launch examples. Under OpenAI's terms, the person who generated an image owns the output.

Those terms are what govern any DALL-E 3 images you kept. OpenAI asked users of the DALL-E GPT to download anything they wanted before August 30, 2026.

What DALL-E 3 cost

DALL-E 3 was priced per image. These were the API rates on OpenAI's pricing page in June 2024 (archived copy):

Standard, 1024x1024: $0.040 per image
Standard, 1024x1792 or 1792x1024: $0.080 per image
HD, 1024x1024: $0.080 per image
HD, 1024x1792 or 1792x1024: $0.120 per image

DALL-E 2 was cheaper at $0.016 to $0.020 per image depending on size. In ChatGPT, DALL-E 3 came with Plus and Enterprise from launch, and from August 2024 Free users got two images a day.

How that compares with today

OpenAI's current image models are billed by tokens, not per image. GPT Image 2.5 Sunburst and GPT Image 2.5 Flare cost $8.00 per million image input tokens ($2.00 cached), $30.00 per million image output tokens and $5.00 per million text input tokens ($1.25 cached), the same rates as GPT Image 2. The Batch API halves that to $4.00, $1.00 and $15.00 for images and $2.50 and $0.625 for text.

For a per-image reference point, the outgoing gpt-image-1 charged $0.011 (low), $0.042 (medium) or $0.167 (high) for a 1024x1024 image. What a 2.5 image costs depends on its size and quality, so run your own prompts through it before you set a budget.

Why OpenAI replaced it

DALL-E 3 was a separate model that ChatGPT called. With GPT-4o, OpenAI built image generation into the chat model itself, and the line has moved fast since:

March 25, 2025: GPT-4o image generation launched in ChatGPT as "our most advanced image generator yet". DALL-E stayed reachable through a dedicated DALL-E GPT.
April 23, 2025: gpt-image-1 brought the new generation to the API, followed by the cheaper gpt-image-1-mini on October 6, 2025.
November 14, 2025: developers were told DALL-E 2 and DALL-E 3 would leave the API.
December 16, 2025: new ChatGPT Images, with gpt-image-1.5 and chatgpt-image-latest in the API.
April 21, 2026: ChatGPT Images 2.0 on all plans and GPT Image 2 in the API, with flexible sizes, high-fidelity image inputs, token pricing and a 50% Batch discount.
May 12, 2026: dall-e-2 and dall-e-3 removed from the API.
August 30, 2026: the DALL-E GPT in ChatGPT retired.
September 8, 2026: ChatGPT Images 2.5, plus GPT Image 2.5 Sunburst and GPT Image 2.5 Flare in the API.

Two paintings of a chocolate chip cookie splashing into milk, a muddy DALL-E 2 version beside a crisp DALL-E 3 version — DALL-E 2 versus DALL-E 3 on the same prompt, from OpenAI's launch material. Its successors have since moved past both.

Each step added something DALL-E 3 could not do: editing from reference images, flexible sizes, C2PA provenance metadata. Once the newer models covered every use case, there was little reason to keep a 2023 model running.

What to use instead

In ChatGPT

ChatGPT Images is the direct replacement and it is on every plan. OpenAI describes the tiers only in words: Free gets "Limited image creation", Go "More image creation", Plus "More complex and accurate image creation" and Pro "Unlimited and faster image creation", subject to abuse guardrails. Image generation with Thinking needs Plus or Pro. Pro costs $100 or $200 a month, and a $500 Pro 500 plan arrived on September 29, 2026. Our ChatGPT review has the full plan breakdown.

In the API

Use gpt-image-2.5-sunburst for "workflows where editing precision matters most", or gpt-image-2.5-flare, which OpenAI calls "our fastest model for high-quality, everyday image generation." Both are available through the Image API and the Responses API image tool. GPT Image 2 remains an option at the same token rates.

Migration tip

Change the model name, then re-test your prompts and sizes. Do not migrate to gpt-image-1, gpt-image-1-mini or gpt-image-1.5: all three retire by December 1, 2026.

Outside OpenAI

Bing Image Creator offered DALL-E 3 next to GPT-4o from August 2025 and is reportedly still a free way to try it, though Microsoft has not confirmed that for 2026. If you are open to other models, we have also reviewed Midjourney, Playground AI and Fooocus, and our Craiyon review covers the old DALL-E Mini lineage.

Verdict

DALL-E 3 was a good model for its time. It made text-to-image usable for people who never wanted to learn prompt syntax, it followed detailed instructions far better than DALL-E 2, and its rights terms were simple. In October 2026 it is a historical product: gone from Azure, from the API and from ChatGPT.

Our rating: no longer recommended, because it is no longer available. If you liked how DALL-E 3 worked inside ChatGPT, ChatGPT Images 2.5 is the same idea with more control. Developers should go straight to GPT Image 2.5 Sunburst or Flare.

Frequently asked questions

Can I still use DALL-E 3?

Not from OpenAI. It was removed from the API on May 12, 2026, and the DALL-E GPT in ChatGPT closed on August 30, 2026. Azure OpenAI retired it on March 4, 2026. Bing Image Creator is the only place it has been reported in 2026, and Microsoft has not confirmed that.

What happens if my app still calls dall-e-3?

The request fails, because the model no longer exists in the API. Switch to gpt-image-2.5-sunburst, gpt-image-2.5-flare or gpt-image-2, and review your size settings and cost estimates, since the newer models support flexible sizes and are billed by tokens.

How much did DALL-E 3 cost?

$0.040 to $0.120 per image in the API. A standard 1024x1024 image was $0.040, while an HD wide or tall image was $0.120. In ChatGPT it was included with paid plans.

Was DALL-E 3 ever free?

Partly. From August 8, 2024, ChatGPT Free users could create up to two DALL-E 3 images per day. Anything more needed a paid ChatGPT plan or pay-as-you-go API billing.

Do I own the images I made with DALL-E 3?

Yes. OpenAI's terms say you own the output and assign you its rights, and the launch page said you could reprint, sell or merchandise DALL-E 3 images without permission. The terms also note that similar output may be generated for other users.

Is gpt-image-1 a safe replacement?

No. gpt-image-1 retires on October 23, 2026, and gpt-image-1-mini and gpt-image-1.5 follow on December 1, 2026. OpenAI points all of them to gpt-image-2.5-sunburst or gpt-image-2.5-flare.

What replaced DALL-E 3 in ChatGPT?

ChatGPT Images, now at version 2.5 since September 8, 2026. It is available on every ChatGPT plan, with more usage and image generation with Thinking on Plus and Pro.

Key takeaways

DALL-E 3 is retired: Azure on March 4, 2026, the OpenAI API on May 12, 2026 and ChatGPT's DALL-E GPT on August 30, 2026.
It made ChatGPT write the prompt for you and beat DALL-E 2 on prompt adherence and in-image text.
API prices were $0.040 to $0.120 per image; today's OpenAI image models are billed by tokens.
Use ChatGPT Images 2.5 in ChatGPT, or gpt-image-2.5-sunburst and gpt-image-2.5-flare in the API.
Skip gpt-image-1, gpt-image-1-mini and gpt-image-1.5 as migration targets: all retire by December 1, 2026.