OpenAI released ChatGPT Images 2.0 on 21 April, powered by the new gpt-image-2 model — the company's first image generation system with native reasoning capabilities built into the architecture. Rather than immediately rendering an image from a text prompt, the model reasons through visual composition decisions before generating output. OpenAI describes the approach as enabling the model to 'search the web for real-time information, create multiple distinct images from one prompt, and double-check its own outputs.' The system supports up to 2K resolution, flexible aspect ratios from 3:1 ultra-wide to 1:3 ultra-tall, and can generate up to eight coherent images from a single prompt with consistent characters and objects maintained across the full set — a capability aimed at storyboarding, manga, and multi-frame design workflows.
The model delivers substantial improvements over DALL-E in areas that matter for professional use: fine-grained text rendering and small typography, accurate UI elements and iconography, tight compositions, and significantly better multilingual support across Japanese, Korean, Chinese, Hindi, and Bengali. Two operating modes ship with the release — Instant for fast generation and Thinking for deliberate, accuracy-focused output. Basic gpt-image-2 generation is available to all ChatGPT users including the free tier, while Thinking mode and advanced reasoning features are restricted to Plus, Pro, and Business subscribers. The model also ships inside Codex, enabling developers to generate images, mockups, and visual assets within the same environment they use for coding — a workflow integration that positions image generation as part of the development pipeline rather than a separate creative tool.
The most consequential detail may be the retirement timeline: DALL-E 2 and DALL-E 3 will be discontinued on 12 May 2026, giving developers and businesses less than three weeks to migrate. For context engineers, gpt-image-2 represents the same reasoning-first pattern that OpenAI has applied across its model family — the model plans before it acts, verifies its output, and uses web search to ground its decisions in real-world information. The Codex integration is particularly relevant for the COR Summit community: developers building product interfaces, documentation, or marketing materials can now generate and iterate on visual assets without leaving their development environment, closing the gap between code and design in a way that tools like Claude Design are also pursuing from the Anthropic side.
