What is DALL·E / GPT Image?

DALL·E was OpenAI’s original text-to-image system, launched in 2021 and folded into ChatGPT with DALL·E 3 in 2023. It made image generation conversational: describe a scene in plain English and ChatGPT would write an optimised prompt and return an image. But it was always a separate diffusion model bolted onto the chatbot.

That era is over. In early 2025 OpenAI replaced DALL·E 3 inside ChatGPT with GPT Image — a natively multimodal model built into the GPT neural network itself, so the same system that understands your words also generates the picture. By mid-2026 the DALL·E models had been removed from the API entirely, leaving the GPT Image family (GPT Image 2 as flagship, 1.5 as the previous generation, and a cheaper Mini) as OpenAI’s image engine. So this review covers both: the DALL·E name people still search for, and the GPT Image reality they’ll actually use.

Key features

Native GPT Image models

The headline change is architectural. GPT Image is autoregressive and integrated directly into the GPT model, rather than a diffusion model called from outside. In practice that means it understands instructions at the same depth as the chat model — which is why it follows complex, multi-part prompts and spatial directions (“put the logo top-left, the text along the bottom”) far more reliably than older systems. It also tops independent quality leaderboards for photorealism and prompt accuracy.

Best-in-class text rendering

Like Ideogram, GPT Image largely solved the text problem that plagues other generators. It renders legible words, headings and labels reliably, making it genuinely useful for posters, product mock-ups, infographics and social graphics where the copy has to be correct — an area where Midjourney still struggles.

Conversational, iterative editing

This is the real standout. Because generation happens inside the conversation, you can refine an image by talking to it: “make the sky darker,” “remove the car,” “change the text to read SALE.” It edits specific regions rather than regenerating from scratch, which turns image creation into an iterative back-and-forth instead of a one-shot gamble — something no standalone generator does as naturally.

Built into ChatGPT, commercial rights included

There’s nothing to install or learn: if you use ChatGPT, you already have it, on web, desktop and mobile. Images you generate are yours to use commercially by default, with no extra licence to buy. For the hundreds of millions of people already in ChatGPT, image generation is simply one more thing the assistant does.

The standout: capability meets convenience

What makes OpenAI’s image generation special isn’t one feature but the stack of them landing in the most convenient place possible. It leads or near-leads on image quality, text rendering and prompt adherence, and it’s the best in the business at conversational editing — all inside the chatbot people already use every day. Where Midjourney asks you to learn its world and Stable Diffusion asks you to build one, GPT Image just works in a conversation. For anyone who wants professional-grade results without adopting a new tool, that combination of raw capability and zero friction is the whole pitch — and it’s a strong one.

Scorecard

Overall score: 8.2 / 10 — the average of the eight categories above.

Pricing

There’s no separate “DALL·E” or “GPT Image” subscription — image generation is bundled into ChatGPT, so your plan decides how much you get. Separately, developers can call the GPT Image models through the OpenAI API on a per-image basis. The free ChatGPT tier includes a small daily allowance; Plus at $20/month is the sweet spot for most image work, with a generous daily quota and full model access.

For everyday creation, if you already pay for ChatGPT Plus, image generation costs you nothing extra — which is a big part of its appeal. On the API, GPT Image 2 (flagship) runs roughly $0.005 to $0.21 per image depending on the quality tier (Low/Medium/High) and resolution, with the cheaper GPT Image 1 Mini from about half a cent; new API accounts get $5 in free credit. It’s worth noting OpenAI’s per-image API rates sit above Google’s Imagen for comparable quality, so very high-volume users should compare. The old DALL·E models are no longer available via the API.

Pros & cons

Honest weaknesses

The clearest limitation is artistic. GPT Image is superb at accuracy, text and editing, but for that distinctive, cinematic, gallery-worthy look Midjourney still has the edge — OpenAI’s output can feel clean and literal rather than inspired. It’s also a feature, not a workshop: there’s no fine-tuning, no LoRA training, no node-based control of the kind Stable Diffusion offers and no production canvas like Leonardo AI‘s, so power users will hit a ceiling.

The content filters are notably strict — reasonable requests can be declined, and named public figures are off-limits — which frustrates editorial and creative users. The branding is genuinely muddled too: the DALL·E name lingers in search results and old tutorials even though the models behind it have been retired, so it’s easy to follow outdated guidance. And while it’s a bargain bundled into ChatGPT Plus, heavy API users will find OpenAI’s per-image pricing higher than some rivals. If guaranteed commercial-safety and Creative Cloud integration matter more, Adobe Firefly is the safer pick.

Who is DALL·E / GPT Image for?

OpenAI’s image generation is the natural choice for ChatGPT users, marketers, small businesses and anyone who wants strong results without learning a dedicated tool — especially for work involving text, accurate prompts or quick iterative edits. If you already pay for Plus, it’s effectively free and excellent. Reach for something else when your priority is pure artistry (Midjourney), deep control and customisation (Stable Diffusion or Leonardo AI), or guaranteed commercial-safety inside a design suite (Adobe Firefly). For capability plus convenience, though, it’s the strongest all-rounder here.

FAQ

AI image generationDALL-EGPT ImageOpenAIAI tool review