GPT Image 2
DALL-E's replacement — an image model that lives inside the same conversation as your text, so it remembers what you just asked for.
What is GPT Image 2?
DALL-E had a genuinely long run for an AI product — five years introducing generative imagery to the mainstream before OpenAI moved on. The shift started quietly in late 2025, when native image generation built directly into GPT-4o began replacing the standalone DALL-E 3 model inside ChatGPT with no formal announcement. That lineage continued through GPT Image 1 and 1.5 before arriving at GPT Image 2, launched April 21, 2026 as "ChatGPT Images 2.0." OpenAI finally closed the loop on the old model on May 12, 2026, permanently shutting down the DALL-E 2 and DALL-E 3 API endpoints for good.
The technical shift matters as much as the branding. Where DALL-E was a diffusion model called as a separate tool, GPT Image is autoregressive and native to the same model that handles your text — meaning it actually remembers what you discussed three messages ago, can reference an image you uploaded earlier, and can refine its own output conversationally rather than starting fresh with every attempt. GPT Image 2 specifically adds a reasoning step before generation, and on release it immediately topped independent leaderboards including LM Arena and Artificial Analysis, reportedly by a record-breaking margin. The trade-off: generation can take a minute or two per image rather than the few seconds a dedicated generator like Midjourney offers, and its default aesthetic leans toward a polished, commercial look that takes explicit prompting to push away from.
Benchmark figures are as reported at GPT Image 2's April 2026 release by independent leaderboards (LM Arena, Artificial Analysis); results shift as competitors release new models. Pricing and features reflect OpenAI's documentation as of July 2026.
Key features
Conversational memory
Understands earlier messages in the chat and can iterate on its own prior output rather than starting over.
Image-to-image editing
Refines or transforms an uploaded or previously generated image based on natural-language instructions.
Reliable text rendering
Places readable, accurate text inside generated images more consistently than most competitors.
Reasoning before generation
Plans out the image before rendering it, a structural change from diffusion-only predecessors like DALL-E.
Instruction-following accuracy
Handles complex, multi-part prompts with more fidelity to specific details than earlier models.
Advanced photorealism
A meaningful step up in realistic rendering compared to the DALL-E 3 generation.
Pricing
ChatGPT Free
- Access to GPT Image within ChatGPT's free tier
- Capped usage, resets daily
- No separate sign-up needed
ChatGPT Plus
- Expanded image generation access
- Included alongside GPT-5.x, browsing & file analysis
- No separate charge per image
API
- $8 per 1M input tokens, $2 per 1M cached inputs
- Pay-as-you-go for developers building on GPT Image
- Pricing varies by output size and quality tier
A ChatGPT Pro tier ($200/month) offers higher usage limits for heavy users. Prices reflect OpenAI's published pricing as of July 2026 and may change — cards link to OpenAI's official pricing page.
Available models
GPT Image is autoregressive rather than diffusion-based, a structural departure from the DALL-E line it replaced.
Integrations & platforms
Pros, cons & best for
Pros
- Conversational iteration makes refining an image genuinely easier
- Included at no extra cost with an existing ChatGPT Plus subscription
- Currently leads major independent image benchmarks
Cons
- Noticeably slower than dedicated generators — up to a minute or two per image
- Default style leans polished and commercial unless you prompt against it
- Old DALL-E 3 workflows and API integrations required a hard migration
Best for
- ChatGPT users who want images without a separate subscription
- Business graphics, presentations and text-heavy designs
- Quick concept mockups refined through conversation
Take a look inside
Alternatives
For faster generation or a more distinctive artistic style:
Our verdict
GPT Image 2 makes the strongest possible case for OpenAI's decision to move on from DALL-E: it's genuinely more capable, tops independent benchmarks, and the conversational memory changes how iteration actually feels compared to firing off isolated prompts. For the tens of millions of people already paying for ChatGPT Plus, it's effectively a free, best-in-class image tool bundled into a subscription they already have. Where it falls short of dedicated generators is speed and stylistic range — if you need results in seconds or a specific artistic aesthetic Midjourney is known for, this isn't the fastest or most distinctive option. As the default image tool for someone already living inside ChatGPT, though, it's hard to beat.
FAQ
What happened to DALL-E 3?
OpenAI phased it out through late 2025 and early 2026, replacing it inside ChatGPT with native GPT-4o and then GPT Image models. DALL-E 2 and 3's API endpoints were permanently shut down on May 12, 2026.
Is GPT Image 2 free?
ChatGPT's free tier includes limited daily image generations. ChatGPT Plus at $20/month includes expanded access at no separate per-image charge, and developers can also pay per image through the API.
How is GPT Image different from DALL-E?
DALL-E was a diffusion model called as a separate tool inside ChatGPT. GPT Image is autoregressive and native to the same model handling your conversation, so it remembers context and can iterate on its own prior output rather than generating in isolation.
Why does GPT Image 2 take longer to generate an image?
It adds a reasoning step before generation, which improves instruction-following and quality but can push generation time to a minute or two per image, slower than dedicated diffusion-based generators.
Can I still use DALL-E 3 through the API?
No, its API endpoints were permanently shut down on May 12, 2026. Any application still calling DALL-E 3 needed to migrate to GPT Image before that date.
Is GPT Image 2 good at rendering text in images?
Yes, it's a relative strength — GPT Image reliably places readable, accurate text inside generated images more consistently than most general-purpose competitors.