Arriving May 2026 · Waitlist open
OpenAI's April 2026 release. 4K generations, 99% text rendering accuracy, and built-in reasoning. We open the playground the day the API goes live.
Released 2026-04-21 · 12 hours to #1 on Image Arena · +242 ELO, largest lead ever recorded
Text that renders
99%+ character accuracy across Latin, CJK, Hindi, and Bengali scripts. Finally reliable for posters, menus, UI mockups.
4K by default
Native 4096×4096 output with crisp details. Roughly 2× faster than the previous generation at the same resolution.
Built-in reasoning
First OpenAI image model with native "thinking" — can search the web for reference, self-check outputs, and plan composition before drawing.
Consistent characters
Generate 8 panels from one prompt with the same character, identical palette, and matching object placement.
Posters & marketing.
Generate print-ready posters with real typography, taglines, and even multilingual variants from one prompt.
UI mockups.
Describe a dashboard and get a pixel-dense screenshot with readable labels, real-looking data, and coherent layout.
Illustration with captions.
Create book covers, comic panels, or educational diagrams where the text is part of the image, not a post-hoc overlay.
OpenAI's latest image generation model (released April 21, 2026). It introduces native reasoning, 4K output, and radically improved text rendering — images can contain readable words, UI copy, logos, and multilingual signage.
Three leaps: (1) text rendering goes from unreliable to ~99% accurate; (2) resolution doubles to 4K; (3) it can plan and revise its output before finalizing (reasoning-first image generation).
OpenAI announced API access arrives in early May 2026. We'll turn on the live playground the day the API opens. Sign up for free to be notified.
Credit-based. A standard 1024×1024 generation is 25 credits. A 4K image is 60 credits. Pro plan includes 1,500 credits/month.
Yes, as long as you have the rights. GPT Image 2 respects the same content policy as OpenAI's other tools — no public-figure likenesses, no copyrighted characters, no safety-blocked content.
Yes — include a reference image and instructions, and the model performs localized edits while preserving the rest. This is best-in-class for product mockups and marketing iteration.
GPT Image 2 took the #1 spot on Image Arena within 12 hours of launch (+242 points, largest lead ever recorded). The specific edge is text rendering + reasoning; open-source models like Flux 2 are still competitive on pure aesthetic but weaker on structured tasks.
Yes. You own the outputs, subject to OpenAI's usage policy. Our terms let you use generated images commercially.