DALL·E 3 is the text-to-image model OpenAI launched in 2023. As of September 2026 it is retired almost everywhere: OpenAI removed it from the API on May 12, 2026, retired the official DALL·E GPT in ChatGPT on August 30, 2026, and Bing Image Creator no longer lists it. OpenAI’s GPT Image models, now at GPT Image 2.5, replaced it.
Last updated: September 2026
What is DALL·E 3?
DALL·E 3 is the third generation of OpenAI’s DALL·E image models. OpenAI announced it in September 2023, built it into ChatGPT for Plus and Enterprise users in October 2023, and opened it to developers through the Images API on November 6, 2023. Microsoft also used it in Bing Image Creator and Microsoft Copilot.
OpenAI said DALL·E 3 understood “significantly more nuance and detail” than earlier versions. Until March 2025, it was the default way people made AI images inside ChatGPT.
On March 25, 2025, OpenAI made GPT-4o image generation ChatGPT’s default image tool. DALL·E 3 stayed available through a dedicated DALL·E GPT and the API, and OpenAI has now closed both.
Is DALL·E 3 still available in 2026?
| Where | Status (September 2026) | What to use instead |
|---|---|---|
| OpenAI API (dall-e-3) | Removed on May 12, 2026; developers were notified on November 14, 2025 | gpt-image-2 or GPT Image 2.5 (Flare or Sunburst) |
| OpenAI API (dall-e-2) | Removed on the same date, May 12, 2026 | Same as above |
| ChatGPT default image generator | Replaced by GPT-4o image generation on March 25, 2025; ChatGPT now runs ChatGPT Images 2.5 (released September 8, 2026) | Ask for an image in any ChatGPT chat |
| Official DALL·E GPT in ChatGPT | Retired on August 30, 2026 (announced July 31, 2026); custom GPTs that generate images are not affected | ChatGPT Images |
| Bing Image Creator | OpenAI’s GPT-4o model was added next to DALL·E 3 in August 2025; the help page now lists only Microsoft’s MAI-Image-2.5-Flash and MAI-Image-2e | Bing Image Creator with MAI models |
| Microsoft Copilot and Microsoft Designer | Both still generate images, but Microsoft’s support and product pages do not name the model | Do not assume they still run DALL·E 3 |
The DALL·E 3 model page in OpenAI’s API docs now reads: “DALL·E 3 has been deprecated and removed from the API. We recommend GPT-Image-2.5 Sunburst for current image generation and editing.”
What replaced DALL·E 3?
OpenAI’s GPT Image family, natively multimodal models that both generate and edit images, replaced it:
- March 25, 2025: GPT-4o image generation becomes the default in ChatGPT.
- April 23, 2025: the model reaches the API as gpt-image-1.
- October 6, 2025: gpt-image-1-mini, a cheaper version, launches.
- December 16, 2025: gpt-image-1.5 launches in ChatGPT and the API.
- April 21, 2026: ChatGPT Images 2.0 launches in ChatGPT, with a “thinking” mode for paid plans; the matching API model, gpt-image-2 (snapshot gpt-image-2-2026-04-21), arrived the same month.
- September 8, 2026: ChatGPT Images 2.5 launches, with two API models: gpt-image-2.5-flare for fast everyday generation and gpt-image-2.5-sunburst for work where editing precision matters most.
The older GPT Image models are being phased out too. On June 2, 2026, OpenAI told developers that gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest will be removed from the API on December 1, 2026, with gpt-image-2 as the replacement. The gpt-image-1 model page is also marked deprecated, although the deprecations page gives no shutdown date for it.
Key features of DALL·E 3
Strong prompt adherence
DALL·E 3 followed long, detailed prompts far more closely than DALL·E 2. GPT Image models build on this, though OpenAI’s guide says they can still struggle to place elements precisely in layout-sensitive compositions.
Readable text inside images
DALL·E 3 rendered short text on signs, posters and labels more reliably than earlier models. OpenAI’s guide still lists precise text placement and clarity as a limitation of GPT Image models, so proofread before you publish.
ChatGPT and GPT-4 rewrote your prompts
In ChatGPT, you described an idea and ChatGPT wrote a detailed prompt for DALL·E 3. According to OpenAI’s cookbook, every DALL·E 3 API prompt was also rewritten by GPT-4, with no way to turn this off. That helped beginners but frustrated developers who needed exact control.
Fixed sizes, quality and style settings
The API offered three sizes (1024×1024, 1792×1024 and 1024×1792), two quality levels (standard and hd), two styles (“vivid” and “natural”) and one image per request.
C2PA content credentials
In February 2024, OpenAI began adding C2PA metadata to DALL·E 3 images from ChatGPT and the API. It warned that metadata is “not a silver bullet” because screenshots and social platforms can strip it. Today, supported OpenAI images carry both C2PA metadata and an invisible SynthID watermark, which you can check at openai.com/verify.
Content policy and artist-style refusals
DALL·E 3 was designed to decline requests in the style of a living artist and requests for public figures by name, and creators could opt their images out of training. The GPT Image API adds a moderation setting: “auto” (default) or “low” (less restrictive).
You own the images
At launch, OpenAI said DALL·E 3 images were yours to reprint, sell or merchandise. Its current Terms of Use, effective January 1, 2026, say you own the output, which also covers GPT Image. Third-party rights such as trademarks and likenesses remain your responsibility.
Benefits of DALL·E 3 and its successors
- Conversational creation: the ChatGPT workflow that DALL·E 3 introduced, where you describe, review and refine in plain language, is now the standard for ChatGPT Images.
- Clear ownership terms: OpenAI’s terms assign output rights to you, which simplifies commercial use.
- Provenance built in: C2PA metadata, and now SynthID, help platforms and audiences identify AI-generated images.
- A free route: ChatGPT Images 2.0 launched on all ChatGPT plans, including the free one, and Images 2.5 rolled out to all ChatGPT users.
Use cases by role
- Marketers and social media managers: campaign concepts, ad variations, social visuals and quick mockups. According to 9to5Mac, OpenAI pitches Sunburst for production-ready campaign creative.
- Founders and small businesses: product scenes, website illustrations and pitch deck images without hiring a designer for every draft.
- Teachers and students: illustrations for lessons, presentations and stories.
- Developers: apps that generate or edit images through the API. Teams that built on dall-e-3 must migrate to a GPT Image model.
- Designers: mood boards, reference images and first drafts to refine in professional tools.
How much does DALL·E 3 cost?
You can no longer buy DALL·E 3. It launched in the API in November 2023 “starting at $0.04 per image” and has not been offered since May 12, 2026. The prices below for its successors are from OpenAI’s pricing and model pages as of September 2026, in USD.
In ChatGPT, image generation is included in every plan, with “images with thinking” reserved for paid plans. In the API, current GPT Image models are billed per 1 million tokens:
| API model | Status | Text input | Image input | Image output |
|---|---|---|---|---|
| gpt-image-2.5-sunburst | Current (September 8, 2026) | $5.00 | $8.00 | $30.00 |
| gpt-image-2.5-flare | Current (September 8, 2026) | $5.00 | $8.00 | $30.00 |
| gpt-image-2 | Current (April 2026) | $2.50 | $4.00 | $15.00 |
| dall-e-3 | Removed May 12, 2026 | — | — | Not available (launched at $0.04 per image and up) |
Cached input costs less ($2.00 per million image tokens for GPT Image 2.5, $1.00 for gpt-image-2) but applies only with the Responses API. OpenAI does not publish fixed per-image prices for these models; it points to the calculator in its image generation guide.
Older GPT Image models list per-image prices, but all three are deprecated, and gpt-image-1.5 and gpt-image-1-mini will be removed on December 1, 2026:
| Model (per image) | Low, 1024×1024 | Medium, 1024×1024 | High, 1024×1024 | High, 1024×1536 |
|---|---|---|---|---|
| gpt-image-1.5 | $0.009 | $0.034 | $0.133 | $0.20 |
| gpt-image-1 | $0.011 | $0.042 | $0.167 | $0.25 |
| gpt-image-1-mini | $0.005 | $0.011 | $0.036 | $0.052 |
Bing Image Creator is free, with 15 fast creations a day and Microsoft Rewards points for more, but it now runs Microsoft’s MAI models, not DALL·E 3.
DALL·E 3 pros and cons
Pros
- Set a new bar for prompt adherence and text in images when it launched in 2023
- Made AI image generation conversational through ChatGPT
- Simple API with predictable per-image pricing
- C2PA provenance metadata from February 2024
- Clear output ownership under OpenAI’s terms
Cons
- Retired: removed from the API on May 12, 2026, and the DALL·E GPT retired on August 30, 2026
- Automatic prompt rewriting in the API could not be disabled
- Only three sizes and one image per API request
- No longer in Bing Image Creator, once the best-known free way to use it
- Its successors bill by token, which makes per-image costs harder to predict for the newest models
How to use DALL·E 3’s replacement today
- For everyday use, open ChatGPT and describe the image in any chat; the retired DALL·E GPT is no longer needed. Refine results with follow-up messages, comment-based edits, templates or the Sketch tool.
- For free generation on Bing, go to bing.com/create and pick an MAI model; expect a different look from DALL·E 3.
- For developers, change the model parameter. Replace dall-e-3 with gpt-image-2.5-flare for fast, high-volume work, gpt-image-2.5-sunburst for precise edits, or gpt-image-2 for the lowest current token prices.
- Update size and quality settings. OpenAI recommends 1024×1024, 1536×1024 or 1024×1536 (custom sizes must be multiples of 16, up to 3,840 pixels on the long edge), and GPT Image 2.5 models take quality values of low, medium, high, xhigh, max or auto instead of standard and hd.
- Check rate limits. DALL·E 3 allowed 500 images per minute at usage tier 1; gpt-image-2 and GPT Image 2.5 Flare allow 5 at tier 1 and 250 at tier 5.
- Test your costs with OpenAI’s image generation calculator before you move production traffic.
DALL·E 3 alternatives
- GPT Image (ChatGPT Images 2.5): the direct successor; pick it if you liked DALL·E 3 in ChatGPT or need an OpenAI API model.
- Midjourney: pick it for stylized, high-quality art and fine aesthetic control; plans run from $10 per month (Basic) to $120 per month (Mega), with 20% off for annual billing.
- Adobe Firefly: pick it for Adobe workflows and models Adobe says are trained on licensed and public-domain content; Firefly Standard costs $9.99 per month.
- Nano Banana (Google Gemini): pick it for free image generation and editing in Gemini, where Nano Banana 2 is Google’s current model.
- Ideogram: pick it for posters, logos and other text-heavy designs; it has a free plan, and Plus costs $15 per month billed annually or $20 monthly.
- FLUX by Black Forest Labs: pick it for an API alternative or open-weight models you can self-host; the current image family is FLUX.2.
Frequently asked questions
Is DALL·E 3 still available in 2026?
No, not from OpenAI. OpenAI removed dall-e-3 and dall-e-2 from its API on May 12, 2026, and retired the official DALL·E GPT in ChatGPT on August 30, 2026. Bing Image Creator’s help page now lists only Microsoft’s MAI models. Image generation in ChatGPT now runs on ChatGPT Images 2.5, OpenAI’s GPT Image model.
When was DALL·E 3 shut down in the API?
On May 12, 2026. OpenAI announced the deprecation on November 14, 2025, and a reminder was posted on OpenAI’s developer forum on April 9, 2026. DALL·E 2 was removed the same day. OpenAI’s deprecations page recommends gpt-image-2, gpt-image-1 or gpt-image-1-mini as replacements, and the DALL·E 3 model page now recommends GPT-Image-2.5 Sunburst.
What replaced DALL·E 3?
OpenAI’s GPT Image family. It started as GPT-4o image generation in ChatGPT on March 25, 2025, and reached the API as gpt-image-1 in April 2025. Later versions include gpt-image-1.5, gpt-image-2 (ChatGPT Images 2.0) and, since September 8, 2026, GPT Image 2.5, offered in the API as Flare and Sunburst.
Can I still use DALL·E 3 in ChatGPT?
No. ChatGPT switched its default image generator to GPT-4o image generation in March 2025 and kept DALL·E 3 only in a dedicated DALL·E GPT, which OpenAI retired on August 30, 2026. OpenAI says custom GPTs that generate images are not affected by the retirement.
Is DALL·E 3 free on Bing Image Creator?
Not anymore. Bing Image Creator is still free, with 15 fast creations a day and Microsoft Rewards points for more, but its help page now lists only Microsoft’s MAI-Image-2.5-Flash and MAI-Image-2e models. For free OpenAI image generation, use ChatGPT, where image generation is included on the free plan.
How much did DALL·E 3 cost, and what does GPT Image cost?
DALL·E 3 launched in the API at $0.04 per image and up. As of September 2026, GPT Image 2.5 costs $30 per million image output tokens and gpt-image-2 costs $15. Deprecated models with per-image prices include gpt-image-1.5, from $0.009 per low-quality 1024×1024 image to $0.20 for high quality at 1024×1536.
Can I use images from DALL·E 3 or GPT Image commercially?
Generally yes. OpenAI’s Terms of Use, effective January 1, 2026, say that, as between you and OpenAI and to the extent the law allows, you own the output, and OpenAI assigns you its rights in it. You are still responsible for not infringing trademarks, copyrights or people’s likenesses.
Verdict
DALL·E 3 mattered. It made AI images follow instructions, put readable text in pictures and made image creation conversational through ChatGPT. But in September 2026 it is a retired product: gone from the API since May 12, 2026, gone from ChatGPT since the DALL·E GPT closed on August 30, 2026, and replaced in Bing by Microsoft’s own models.
If you came here to use DALL·E 3, use ChatGPT Images 2.5 instead; it is the direct successor and is available on the free plan. Developers should migrate to gpt-image-2.5-flare, gpt-image-2.5-sunburst or gpt-image-2, and should check rate limits and token-based costs before switching. For different styles or pricing, Midjourney, Adobe Firefly, Nano Banana, Ideogram and FLUX are all strong options.
Sources
- Deprecations (OpenAI API docs)
- DALL·E 3 model page (OpenAI API docs)
- Deprecation Reminder: DALL·E will be shut down on May 12, 2026 (OpenAI Developer Community)
- API pricing (OpenAI)
- GPT-Image-2.5 Flare model page (OpenAI API docs)
- GPT-Image-2.5 Sunburst model page (OpenAI API docs)
- GPT-Image-2 model page (OpenAI API docs)
- GPT-Image-1.5 model page (OpenAI API docs)
- GPT-Image-1 model page (OpenAI API docs)
- GPT-Image-1 Mini model page (OpenAI API docs)
- Image generation guide (OpenAI API docs)
- Introducing ChatGPT Images 2.5 (OpenAI Developer Community)
- Introducing ChatGPT Images 2.5 (OpenAI)
- ChatGPT release notes (OpenAI Help Center)
- DALL·E 3 (OpenAI)
- New models and developer products announced at DevDay (OpenAI)
- What’s new with DALL·E 3? (OpenAI Cookbook)
- Introducing 4o Image Generation (OpenAI)
- Provenance signals (Content Credentials, SynthID) in OpenAI-generated content (OpenAI Help Center)
- Terms of Use (OpenAI)
- AI images generated on DALL-E now contain the Content Authenticity tag (PetaPixel)
- OpenAI releases ChatGPT Images 2.5 with ‘sharper details’ and ‘more precise editing’ (9to5Mac)
- GPT Image (Wikipedia)
- DALL-E (Wikipedia)
- Help (Bing Image Creator)
- Bing Image Creator adds GPT-4o alongside DALL-E3 for even more free ways to create (Microsoft Bing Blogs)
- Using image generation in Microsoft Copilot (Microsoft Support)
- Microsoft Designer
- Comparing Midjourney Plans (Midjourney docs)
- Adobe Firefly plans and pricing (Adobe)
- Nano Banana 2 image generation (Google Gemini)
- Nano Banana image generation (Gemini API docs)
- Ideogram pricing
- FLUX models (Black Forest Labs)
