OpenAI just killed DALL-E. Today they officially launched ChatGPT Images 2.0, the successor model that makes DALL-E 3 obsolete overnight. DALL-E 2 and DALL-E 3 are being retired on May 12. The model everyone has used for the last three years to generate AI images is getting pulled, replaced by something OpenAI is calling the first image model that actually thinks. The new model, gpt-image-2, is rolling out to all ChatGPT and Codex users today. Plus, Pro, and Business subscribers get the full thinking capabilities. It is already available via API, on fal, and in Microsoft Foundry. And based on what it can actually do, calling it a DALL-E successor is understating it. This is a complete rebuild of how AI generates images.

OpenAI is not calling this a rendering model. They are calling it a visual thought partner. The difference is that Images 2.0 reasons through complex visual tasks before generating, verifies its own outputs, and can produce up to 8 coherent images from a single prompt. It is the first image model with native thinking capabilities built in.

Here is what actually changed.

Perfect text rendering. This is the big one. Every AI image model before this one had the same catastrophic weakness. Try to generate a Mexican restaurant menu and you got "enchuita," "churiros," "burrto," and "margartas." Try to make a product label and you got gibberish. Try to build a UI mockup and you got warped lettering. Images 2.0 solves it. The model now renders legible, correctly spelled text inside images including signage, product labels, UI mockups, and handwritten notes. Across Latin scripts. Across Japanese, Korean, Chinese, Hindi, and Bengali. Near perfect.

2K native resolution. Standard output resolution is now 2K. The API supports up to 4K. That means the images are now useful for actual production work. Professional product photography. Magazine layouts. Marketing assets. Design mockups. Anything that needs to go into print or high resolution digital just became possible without post processing.

World aware photorealism. The model understands physics, lighting, and material properties. The warm orange color cast that defined every GPT Image 1.5 output is gone. Complex multi object scenes render with correct occlusion and placement. Shadows fall in the right direction. Reflections match the environment. The images look like photographs rather than renders.

8 images from one prompt. Submit one prompt, get up to 8 coherent variations. Not 8 random images. 8 related designs that work as a set. Product shots from different angles. Ad variants with consistent branding. Social campaign assets that feel like they belong together.

Native reasoning. This is what makes it feel different. Before generating, the model works through the prompt. It understands briefs. It makes creative decisions. Canva's creative strategist described it by saying "the model was not just rendering images, it was interpreting briefs, understanding audiences, and making creative decisions behind the scenes."

Updated knowledge cutoff. December 2025. That matters more than it sounds. Image models trained on older data produce outdated references, wrong logos, and incorrect product designs. A December 2025 cutoff means accurate visual context for current events, current products, current design trends.

Because OpenAI was losing the image game and just reclaimed it in one launch.

Before today, Google's Gemini held first place on the LM Arena text to image leaderboard. OpenAI's gpt-image-1.5 was stuck in second. Midjourney, Stability AI, and Ideogram were all pushing hard on specialty use cases. The narrative had shifted. OpenAI was the chatbot company. The image generation title was slipping away.

Today that changed.

Images 2.0 does not just match the competition. It pulls ahead on the one thing every other model has failed at for three years. Text rendering has been the persistent failure point of AI image generation since the beginning. Every major model shipped promising better text and delivered slightly less broken gibberish. Images 2.0 is the first model that actually solves it. That alone makes it production ready for an enormous category of commercial work that was previously impossible.

And the retirement of DALL-E 3 is the signal OpenAI wanted to send. They are not positioning this as an alternative or an upgrade. They are replacing the old product entirely. DALL-E 2 and DALL-E 3 are being retired on May 12. After that, if you want OpenAI image generation, you use Images 2.0 or you use nothing.

⚡ The Vibe Check: Three weeks ago Anthropic dropped Claude Design. Seven days ago Claude Opus 4.7 launched. Four days ago Claude Design shipped and Figma stock fell 7 percent. Today OpenAI killed DALL-E and relaunched the entire image generation category. The pace of this industry is not slowing. It is compounding. Every week this year has had a launch that would have defined an entire quarter in 2024.

If you generate images with AI, your default just changed. Images 2.0 is available in ChatGPT and Codex starting today. Free users get basic access. Plus, Pro, and Business subscribers get the thinking capabilities. If you build with image generation, the API is live. Pricing varies by quality and resolution. fal and Microsoft Foundry are already deployed.

If you work in design, marketing, or content creation, the text rendering fix changes what is possible. Menus, flyers, product labels, social graphics, UI mockups, and advertising assets that used to require Photoshop clean up now work out of the box. A small creative team just got the output capacity of a much larger one.

If you are watching the AI race, here is what to pay attention to. OpenAI was losing on image generation. They just won it back. Google still leads in multimodal embeddings and search integration. Anthropic leads in coding and knowledge work. Every major lab is now clearly ahead in at least one category. The era of one AI company dominating everything is over. The era of specialized frontier leaders is here.

The biggest question is what OpenAI drops next. GPT-5.5 Spud has been in safety evaluation since March 24. Altman said a few weeks away two weeks ago. If Images 2.0 is the image model launch, Spud is the language model launch that follows. This week is not done.

OpenAI killed DALL-E today. The replacement is Images 2.0 powered by gpt-image-2, the first image model with native reasoning, 2K resolution, perfect text rendering, and the ability to generate 8 coherent images from one prompt. DALL-E 2 and DALL-E 3 are retiring May 12. The model is rolling out to all ChatGPT and Codex users starting today. It is available via API, on fal, and in Microsoft Foundry. Text rendering, the single biggest weakness of every AI image model for three years, is finally solved. OpenAI was losing the image generation race. They just reclaimed it. And they did it one week after Anthropic's biggest product launch of the year. Every week in AI right now is the most important week in AI.

How To Use GPT Image 2 👉 Here