DALL-E is effectively a retired brand. OpenAI removed DALL·E 2 and DALL·E 3 from its API on May 12, 2026, and the current image stack is the GPT Image family — GPT Image 2 as the flagship, GPT Image 1.5 below it, GPT Image 1 scheduled for its own deprecation in October 2026, and GPT Image 1 Mini as the budget option. API pricing runs from roughly half a cent per image on Mini up to around twenty-one cents at the flagship's highest quality tier, with batch processing available at a substantial discount. For most people, though, none of this matters: image generation is simply a built-in feature of ChatGPT, included on the free tier with limits and unrestricted on Plus at $20 a month.
Reviewed by EdGrows·Updated Aug 24, 2026Affiliate link — we may earn a commission
This analysis is based on documentation, public user reports, and vendor materials — not yet on our own hands-on testing. How we rate
Type "DALL-E pricing" into a search box and you'll get a lot of confident articles. Most of them are describing something that no longer exists.
OpenAI removed DALL·E 2 and DALL·E 3 from its API on May 12, 2026. The documentation had flagged them as deprecated well before that. What runs today is the GPT Image family, and there is no path to building anything new on DALL-E because the endpoints are gone.
The name persists anyway, for the same reason people still say Hoover. DALL-E was the first AI image generator most people encountered, and it turned into shorthand for the whole category. Search volume for the retired brand still dwarfs volume for its replacement, which is exactly why the internet is full of 2026 articles titled after a 2023 product.
| Model | Per-image range | Status |
|---|---|---|
| GPT Image 2 | ~$0.005–$0.211 | Current flagship |
| GPT Image 1.5 | ~$0.009–$0.20 | Previous flagship, still solid |
| GPT Image 1 | ~$0.011–$0.25 | Deprecating October 23, 2026 |
| GPT Image 1 Mini | ~$0.005–$0.052 | Budget tier |
Each model offers low, medium and high quality tiers, and prices vary with resolution across square, portrait and landscape formats. The wide range within each model is that quality-and-size spread, not a volume discount.
Two things worth flagging for developers. First, GPT Image 1 has its own deprecation date coming in October, so if you migrated off DALL-E onto it, you're due another migration. Second, OpenAI's batch API cuts costs substantially for work that can tolerate delay — that's the largest single lever available on a high-volume image bill, and it costs nothing but scheduling.
The API pricing above is relevant to maybe five percent of the people reading this page. Everyone else generates images by asking ChatGPT for one.
That's included. Free tier users get image generation with rate limits they will notice. Plus at $20 a month removes the constraints for normal use. Pro at $200 exists for heavy workloads.
And this is genuinely the product's strongest argument. You don't learn a tool, manage a separate subscription, or think about models and quality tiers. You're already in ChatGPT for something else, you describe a picture, you get one. The friction is close to zero and no competitor matches that, because no competitor is already open in your browser.
Being clear about this matters more than the pricing, because a lot of coverage treats OpenAI's image generation as a leader when it isn't.
Prompt adherence is the real strength. Give it a complicated instruction with multiple specific requirements and it follows more of them, more literally, than most competitors. If you need a particular thing in a particular arrangement, this is where it wins.
Text inside images works. Most generators still produce garbled lettering. OpenAI's models handle words in images noticeably better, which matters for mockups, posters and anything with a sign or label in it.
Aesthetic quality lags. Midjourney produces better-looking images. Not marginally — consistently, across styles, in a way that's obvious in side-by-side comparisons. If the picture's job is to look good rather than to contain specific elements, that gap is the whole decision.
Style consistency across a set is weak. Generating twelve images that feel like they belong together is hard here and much easier in tools built for it.
Control is limited. No seeds to work from, limited fine-grained direction, minimal reproducibility. Stable Diffusion and its ecosystem exist because some people need that control, and OpenAI does not offer it.
We haven't run comparative generation tests across models for this page, and image quality claims are the kind of thing that deserves that caveat loudly.
What this is based on: OpenAI's published model documentation and deprecation notices for the retirement dates and model lineup, cross-referenced against independent API pricing trackers published between March and August 2026 — the most recent from within the past month. The May 12 removal date and the October 23 deprecation for GPT Image 1 are both documented rather than inferred.
The research difficulty on this topic is unusual and worth naming. Because "DALL-E" carries enormous search volume, a large share of the pages ranking for it are optimized for the old name while describing new products, sometimes inconsistently within the same article. We found current-dated pages quoting DALL·E 3's per-image rate as though it were purchasable. Some of that is honest lag, some of it is SEO. Either way, treat any DALL-E pricing figure with suspicion unless the page states which endpoint it's describing.
Our quality assessments above reflect the consensus in comparative coverage rather than our own benchmarking, and that's a weaker form of evidence than testing would be.
Against Midjourney: better images, worse convenience. Midjourney's aesthetic quality is its whole proposition and it delivers. You pay a separate subscription and learn a separate workflow. For anyone whose output quality matters visually, that trade is worth making.
Against Stable Diffusion: open weights, local execution, unlimited control, no per-image cost after hardware. Also a real learning curve and a GPU requirement. The right answer for people who want to tinker or who can't send prompts to a third party.
Against Flux AI: strong quality with more open licensing options, and a favorite among developers who want something better than OpenAI without Midjourney's closed ecosystem.
Against Ideogram: the specialist on text in images, which is one of the few areas OpenAI does well. If your work is posters, logos and typography-heavy design, worth comparing directly.
Against Adobe Firefly: the commercially safest option, trained on licensed content, integrated into Creative Cloud. For agencies and brands where provenance is a procurement question, that matters more than a quality comparison.
Against Recraft: built for design work specifically — vectors, brand consistency, sets of assets that match. A different job than general image generation.
Against Gemini: the closest structural comparison, since Google also bundles image generation into an assistant people already use. Google's image models have been competitive and the pricing per image runs lower at the budget end.
| Route | Cost | Notes |
|---|---|---|
| ChatGPT Free | $0 | Image generation included, rate limited |
| ChatGPT Plus | $20/mo | Practical limits removed for normal use |
| ChatGPT Pro | $200/mo | Heavy workloads |
| API — GPT Image 2 | ~$0.005–$0.211/image | Current flagship, 3 quality tiers |
| API — GPT Image 1.5 | ~$0.009–$0.20/image | Previous flagship |
| API — GPT Image 1 Mini | ~$0.005–$0.052/image | Budget option, competitive at low/medium |
| Batch API | Substantial discount | For work that tolerates delay |
Checked August 2026. DALL·E 2 and DALL·E 3 were removed from OpenAI's API on May 12, 2026 and are no longer callable. GPT Image 1 is scheduled for deprecation on October 23, 2026. Per-image prices vary by quality tier and resolution; ranges above reflect that spread. Verify current rates on OpenAI's pricing page before budgeting production workloads.
Don't build on DALL-E. There's nothing to build on. If a tutorial tells you to set the model parameter to dall-e-3, it's out of date.
Check what GPT Image 1 integrations are still live. October's deprecation will break anything still pointing at it.
Use batch for anything that isn't interactive. The discount is significant and the only cost is latency.
Try Mini before assuming you need the flagship. At low and medium quality it's competitive, and it's an order of magnitude cheaper at the top end of the range.
OpenAI's image generation is the most accessible in existence and roughly the fourth-best. Both halves of that sentence are true and the tension between them is the entire evaluation.
The DALL-E era is over as a product decision, not just a naming one. The API endpoints are gone, the successor family has its own internal deprecation schedule, and anyone maintaining an integration should assume another migration is coming. That churn is a real cost of building on OpenAI's image stack.
Convenience is the actual product. Being inside ChatGPT means zero setup, zero extra subscription and zero learning curve. For the enormous number of people who want a picture occasionally and don't care about the last twenty percent of quality, that's decisive and no amount of benchmark superiority changes it.
Quality is competent, not leading. Midjourney looks better. Stable Diffusion offers more control. Firefly is safer commercially. Ideogram handles typography better. OpenAI wins on prompt adherence, on text rendering, and on already being open in your browser.
For casual users and anyone already paying for ChatGPT: this is free capability you should use, and there's no reason to add a second subscription unless the output quality is failing you. For designers, marketers and anyone whose images represent a brand: the tools built for that job are better at it, and the gap is visible enough to justify the extra cost. For developers: the GPT Image API is well-integrated and reasonably priced, and the deprecation cadence is a genuine consideration worth planning around.
Note: AIVario earns no commission from OpenAI. This page is based on published documentation and independent pricing research rather than comparative generation testing.
Best for: Casual image generation, existing ChatGPT subscribers, prompts requiring literal adherence to complex instructions, images containing readable text Not ideal for: Professional design work, style-consistent image sets, anyone needing seeds and reproducible control, brand work where training-data provenance matters Bottom line: A retired brand attached to a competent successor whose real advantage is that it's already where you are. Good enough for most people, not good enough for people who care.
As an API product, yes. OpenAI removed DALL·E 2 and DALL·E 3 from the API on May 12, 2026, and its documentation had marked them as deprecated for some time before that. The name persists in search traffic, in older tutorials, and casually among people who say DALL-E when they mean OpenAI image generation. What actually runs today is the GPT Image family. If you're building anything new, there is no version of DALL-E to build it on.
The GPT Image models. GPT Image 2 is the current flagship with the strongest prompt adherence and photorealism. GPT Image 1.5 sits below it and remains solid for production work where the flagship's quality isn't required. GPT Image 1 is the earlier flagship and is itself scheduled for deprecation on October 23, 2026, so any integration still pointing at it needs migrating. GPT Image 1 Mini is the budget tier, considerably cheaper and competitive at lower quality settings.
It depends on model, quality tier and resolution. Mini runs from about half a cent per image up to roughly five cents. GPT Image 1.5 spans roughly a cent to twenty cents. GPT Image 2 covers about half a cent to twenty-one cents depending on settings. Each model offers low, medium and high quality tiers across square, portrait and landscape resolutions. OpenAI's batch API can cut costs substantially for work that doesn't need to happen immediately, which is the single biggest lever on a high-volume image bill.
No, though the free tier is limited. Image generation is built into ChatGPT and free users get access with rate limits that most people will notice. ChatGPT Plus at $20 a month removes the practical constraints for typical use, and Pro at $200 covers heavy workloads. For casual image generation, the free tier genuinely works. For anyone using images regularly, Plus is the sensible tier and the images are a bonus on top of everything else that subscription includes.
On raw image quality, generally no. Midjourney remains stronger on aesthetics, style consistency and the kind of output people actually want to look at. Where OpenAI's models win is prompt adherence — they follow complicated instructions more literally — and text rendering inside images, which most generators still handle poorly. The honest positioning is that OpenAI's image generation is convenient and competent rather than best-in-class, and its main advantage is that it's already inside a tool you're using for other reasons.
OpenAI's terms grant users ownership of outputs generated through its services, subject to the usage policies, so commercial use is generally permitted. That said, the broader legal picture around AI-generated imagery — training data, copyright registration, downstream liability — remains genuinely unsettled and varies by jurisdiction. For casual or internal use this is a non-issue. For anything going into a funded advertising campaign or a product you're selling, get an actual opinion from someone qualified rather than relying on a tool review.
Because the name did its job. DALL-E was the first AI image generator most people heard of, and it became a generic term the way Hoover did for vacuum cleaners. Search volume for DALL-E vastly exceeds volume for GPT Image, which is why so many current articles are titled after a product that no longer exists. It's worth knowing this when you're reading comparisons: a 2026 piece about DALL-E pricing may be describing genuinely current OpenAI image pricing, or it may be quoting a retired model's rate card. Check which.
The leading AI voice platform — text-to-speech, voice cloning, and conversational AI in 70+ languages.
The design suite everyone already uses, now bundled with a free professional toolset — and a credit system worth understanding first.
Category-defining design tool with unusually smart seat pricing — provided you assign the right seat type to each person.