OpenAI launched GPT Image 2 — branded to consumers as ChatGPT Images 2.0 — on April 21, 2026, positioning it as a generational leap over GPT Image 1.5 rather than an incremental update. The headline spec is native 4K output: landscape wallpapers, 1792×1024 cinematic frames, and 3200×1800 infographics that previously required an upscaling pass now come out of a single generation call. Alongside the launch, OpenAI confirmed DALL-E 2 and DALL-E 3 are both deprecated, with full retirement on May 12, 2026 — closing the book on the models that first put AI image generation in front of a mainstream audience back in 2022.
Watch: How to Use ChatGPT’s 4o Image Generator (Real-World Examples) (YouTube)
The number that actually matters more than resolution
Native 4K is the headline, but the bigger practical shift is text rendering: GPT Image 2 claims 99% text accuracy, up from what earlier models managed with basic word rendering that frequently garbled anything longer than a couple of characters. Legible, reliable text has been the single most consistent failure mode across every major image model to date — it’s the reason AI-generated posters, packaging mockups, and UI comps have needed manual text replacement in post. Closing that gap, if the claimed accuracy holds up in practice, moves image generation from “great for concept art, unusable for anything with a label” to something closer to production-ready for a much wider range of commercial work.
Speed and consistency, not just resolution
Generation time drops to roughly three seconds, down from 8–18 seconds on the previous model — a difference that matters enormously for iterative creative workflows where a person is regenerating variations dozens of times per session rather than committing to one prompt. GPT Image 2 also generates up to eight coherent images from a single prompt while maintaining consistent characters and objects across the full set, addressing another long-standing pain point: keeping a character’s face, an object’s design, or a brand’s visual language stable across a multi-image sequence without a separate fine-tuning step.
Pricing lands lower, not higher
Despite the capability jump, 1024×1024 high-quality generation actually drops in price, from roughly $0.25 to $0.211 per image, with 2K and 4K output metered by token in a way that trends toward lower long-term cost rather than a premium surcharge for the added resolution. Combined with the DALL-E retirement forcing a hard migration deadline, it’s a release clearly aimed at consolidating OpenAI’s image-generation user base onto one current model rather than maintaining a legacy tier alongside it.