Introducing gpt-image-2 - available today in the API and Codex

gpt-image-2 is OpenAI’s most capable image generation model yet: a state-of-the-art image model that can take on complex visual tasks and produce precise, immediately usable visuals with sharper editing, richer layouts, stronger text rendering, and thinking-level intelligence.

YouTube

Starting today, developers can build with gpt-image-2 in the API, and use it in Codex to turn ideas, product context, and working materials into polished visual assets.

This release is aimed at production workflows: the moments where an image needs to be more than interesting. It needs to be accurate, readable, on-brand, localized, formatted for the surface where it will ship, and useful without a long cleanup pass.

Independent evaluations already confirm the leap

Just hours after launch, gpt-image-2 claimed the #1 spot across all Image Arena leaderboards — a clean sweep with a record-breaking +242 point lead in Text-to-Image (the largest gap the arena has ever seen), plus dominant wins in Single-Image Edit and Multi-Image Edit. No model has ever led by margins this wide.

What’s new

Create assets for the surface you need

gpt-image-2 supports many more export ratios and higher-resolution outputs up to 2K, making it easier to create images for apps, ads, product flows, social placements, presentations, documentation, and internal tools.

Text-heavy visuals are more practical

The model is stronger at structured image generation, including diagrams, infographics, charts, posters, comics, multi-panel scenes, and other visuals where layout and typography matter. It also improves multilingual text rendering, including non-Latin scripts.

Better control from prompt to final asset

gpt-image-2 is better at following detailed instructions, preserving requested details, relating objects accurately, and rendering dense compositions. That means less “almost right” and more outputs you can actually use.

Thinking mode unlocks richer visual work

When used with a reasoning model, ChatGPT Images 2.0 can research, reason, transform source materials, generate multiple distinct images from one prompt, and check its own work. This makes it especially useful for builders creating assets that depend on current context, product details, or real-world information.

Model capability gallery

The official gpt-image-2 release blog was illustrated entirely with images generated by the model itself. Every single visual, from the very first frame to the final detail, was born from gpt-image-2. No external assets. No stock photography. Just the model showing the world exactly what it can do.

Here we’re sharing a selection of those very images from the blog — each one highlighting a demonstration of the model’s capabilities.

Start Creating

• Experiment instantly in the Image Playground

• Dive into the full Image Generation Guide for tips, code samples, best practices, and the cost calculator with gpt-image-2

• Pricing (per 1M tokens) : Image modality : $8.00 input / $2.00 cached input / $30.00 output Text modality : $5.00 input / $1.25 cached input / $10.00 output Full details and rate limits are on the gpt-image-2 model page

Use gpt-image-2 in the API for production image generation workflows, or try it in Codex when you want to create visual assets from the context of what you are already building.

We’re genuinely excited to see what you build with it. If you create something useful, surprising, beautiful, practical—or all of the above—share it here in the forum. We love seeing your work and hearing how it’s making your workflows better.