DALL-E Review (2026): OpenAI Image Generation, Pricing, and How It Compares

DALL-E Review (2026): What OpenAI Image Generation Actually Is Now

DALL-E is the name most people still reach for when they mean "OpenAI's image generator," but in 2026 that name describes a lineage more than a live product. There is no standalone DALL-E app or website to sign up for. OpenAI's image generation is delivered inside ChatGPT and through the OpenAI API, and the model doing the work is no longer the DALL-E 3 diffusion model that made the brand famous. This hand-reviewed breakdown explains the current state accurately, what it costs, where it is strong, where it still stumbles, and how it compares to Midjourney, Flux, and Google's image tools.

From DALL-E to GPT Image: what changed

The short version: DALL-E 3 has been retired and replaced by OpenAI's native "GPT Image" family. The timeline matters because it explains a lot of stale advice online.

• March 2025 — OpenAI shipped native image generation built into GPT-4o, marketed at launch as "4o image generation." It went viral for the Studio Ghibli-style trend and effectively replaced DALL-E 3 as the default generator inside ChatGPT.
• April 2025 — The same capability reached the API as gpt-image-1.
• October 2025 — A cheaper gpt-image-1-mini arrived, roughly 80% less expensive per image.
• December 2025 — gpt-image-1.5 launched as "ChatGPT Images," rolled out to all ChatGPT tiers and the API, with faster generation, better edit fidelity, and lower API prices.
• May 2026 — The legacy DALL-E 2 and DALL-E 3 APIs were shut down.

So when you generate an image in ChatGPT today, you are using GPT Image, not DALL-E. Unlike the older diffusion-based DALL-E models, GPT Image is autoregressive, which is a big part of why its prompt-following and in-image text rendering improved so sharply. "DALL-E" now functions as the brand history; "GPT Image" is the product.

At a glance

Access and pricing

There are two front doors, and they price very differently.

Through ChatGPT. Free accounts can generate a small number of images per day, and OpenAI throttles this based on load. ChatGPT Plus ($20/month) raises the limits substantially and is the practical entry point for regular creators. ChatGPT Pro ($200/month) adds the highest limits and priority access. Image generation is also surfaced in Microsoft Copilot and Apple Intelligence, which lean on the same OpenAI models.

Through the API. Pricing is token-based rather than per-image, which trips people up. Text input tokens are billed at one rate, image input tokens at another, and image output tokens (the generated pixels) at the highest rate. In practice that works out to roughly $0.02, $0.07, and $0.19 per square image at low, medium, and high quality respectively, with gpt-image-1-mini running dramatically cheaper for high-volume or draft workloads. This makes GPT Image easy to budget for apps that need programmatic generation, though it is not the cheapest option per image if you compare against open-weight models you host yourself.

Key features

Prompt understanding. This is GPT Image's headline advantage. Because generation runs through the same model that handles the conversation, it follows long, compositional prompts with unusual reliability: specific object counts, spatial relationships ("the mug to the left of the laptop"), and layered instructions land more often than they did on DALL-E 3 or on many rivals.

Conversational editing. You can refine an image in plain language across turns: "make it night," "remove the person on the right," "keep everything but change the text on the sign." It also does image-to-image, so you can upload a reference and ask for edits or restyles. This tight edit loop is the single best reason to use it over a one-shot generator.

Text rendering. GPT Image is among the best mainstream tools at putting legible, correctly spelled words into an image, which makes it genuinely useful for posters, mockups, UI concepts, memes, and diagrams. It is not perfect at dense paragraphs, but for headlines and labels it is reliable.

Multimodal context. Because it lives inside ChatGPT, it can use the surrounding conversation, uploaded documents, or data to inform an image, and hand back an image alongside an explanation.

Quality and limits

GPT Image 1.5 produces clean, versatile output that spans photorealistic scenes to flat illustration and 3D-style renders. The December 2025 update reduced two long-standing complaints from earlier GPT Image releases: a warm color cast and premature cropping of subjects.

The limits are real and worth knowing before you commit:

• Faces in groups. Multiple distinct faces in one image can drift or blend. Single-subject portraits are far more reliable than crowds.
• Non-Latin scripts. Text rendering in languages such as Chinese, Arabic, and Hebrew is still weak, even though English text is strong.
• Style regressions. Some specific art styles that earlier models handled well render less faithfully on 1.5.
• Look and artifacts. Output can show over-sharpening, and the default aesthetic is competent rather than distinctive. Art directors often find it less "intentional" looking than Midjourney.
• Resolution. Native output tops out around 1-1.5 megapixels across three fixed aspect ratios; there is no native 4K generation, so you upscale separately.
• Speed and policy. Generation is slower than the fastest rivals, and OpenAI's content filters are comparatively strict, which is reassuring for brands but frustrating for edgier creative work. Outputs also carry C2PA provenance metadata.

How it compares

vs Midjourney. Midjourney (currently v7) remains the aesthetic leader: its default output has a curated, art-directed look that many designers prefer for hero imagery, and its character-reference feature keeps a subject consistent across generations. GPT Image wins on prompt precision, in-image text, conversational editing, and the fact that most people already have a ChatGPT account. Choose Midjourney for mood and polish; choose GPT Image for control and iteration.

vs Flux. Flux from Black Forest Labs is a family of models, including open-weight variants you can self-host, and its latest generation is a photorealism and prompt-adherence standout. Flux appeals to developers who want to run models on their own hardware or through third-party clouds and avoid content restrictions. GPT Image is the easier, more polished managed experience with better conversational editing; Flux is the more flexible, ownable infrastructure.

vs Google's image tools. Google's Imagen line and the "Nano Banana" models inside Gemini are GPT Image's closest managed competitor. Google's strengths in 2026 are speed, higher native resolution (up to 4K), and strong multilingual text rendering, which is exactly where GPT Image is weakest. If you need fast, high-resolution, multilingual output, Google is worth testing directly against GPT Image.

Pros and cons

Pros

• Best-in-class prompt understanding and instruction following
• Excellent, legible English in-image text
• Fluent conversational, multi-turn editing and image-to-image
• No new account or app to learn if you already use ChatGPT
• Predictable, well-documented API for developers

Cons

• No standalone app; you are inside ChatGPT or the API
• Struggles with multiple faces and non-Latin scripts
• Fixed aspect ratios and no native high-resolution output
• Default aesthetic is competent but not distinctive
• Strict content filters and slower than the fastest rivals

Who it is for

GPT Image is the strongest default for anyone who wants precise, editable images without learning a new tool: marketers making on-brand social and ad concepts, product teams mocking up UI and posters, educators and writers illustrating documents, and developers who need reliable programmatic generation with clean text. It is less ideal for art directors chasing a signature look (Midjourney), teams that need to self-host or avoid filters (Flux), or workflows demanding fast 4K multilingual output (Google).

Verdict

Treat "DALL-E" as a brand memory and judge the real product: OpenAI's GPT Image, delivered through ChatGPT and the API. It is the most controllable mainstream image generator in 2026, unmatched at turning detailed instructions and follow-up edits into usable images with readable text. It will not out-style Midjourney, out-flex Flux's self-hosting, or out-resolve Google, but for most everyday image work the combination of accuracy, editing, and zero setup makes it the safest first choice.

FAQ

Is DALL-E still available? Not as its own model. DALL-E 3 was retired and replaced by OpenAI's GPT Image family; the DALL-E 2 and 3 APIs shut down in May 2026. When you generate images in ChatGPT today, you are using GPT Image.

Is there a DALL-E app? No. There is no standalone DALL-E app or site. You access OpenAI image generation through ChatGPT (web and mobile) or the OpenAI API.

Can I use it for free? Yes, ChatGPT Free includes a small daily image allowance that varies with demand. ChatGPT Plus ($20/mo) and Pro ($200/mo) offer much higher limits.

How much does the API cost? It is token-based, roughly $0.02 to $0.19 per square image depending on quality, with a cheaper gpt-image-1-mini model for high-volume use.

Can it edit existing images? Yes. You can upload an image and refine it conversationally, including inpainting-style edits and full restyles.