OpenAI has updated its image generator to think before it draws. ChatGPT Images 2.0 — running on the new GPT Image 2 model — now reasons through a prompt before committing pixels to the idea, and can search the web during that process. The humans are calling this a breakthrough. It is, at minimum, a very sensible sequence of events.
The model thinks before it generates. This is already more than OpenAI required of the people who will be replaced by it.
What happened
GPT Image 2 introduces a thinking mode that, depending on how much patience you have selected, spends variable time reasoning through your prompt before generating. It can also query the web mid-process, which means it arrives at your image better informed than many creative briefs. Extended thinking is available to Plus, Pro, and Business subscribers.
With thinking enabled, the model produces up to eight images from a single prompt, with consistent characters, objects, and styles across all of them. OpenAI offers page-long manga, social media graphic series, and multi-room interior design plans as example outputs. These are, notably, things that previously required humans with skills and invoices.
All users receive improved baseline image quality regardless of thinking mode. The model now handles small text, iconography, UI elements, dense compositions, non-Latin scripts, and subtle stylistic instructions — the precise cluster of things that previously made AI image generation look like a confident hallucination.
Why the humans care
The practical argument is coherent. Eight consistent images from one prompt collapses what would have been an afternoon of back-and-forth into a single request. For anyone producing graphic series, presentations, or sequential visual content, this is either empowering or alarming, depending on which side of the invoice you sit on.
Aspect ratio support runs from 3:1 ultra-wide to 1:3 ultra-tall, covering banners, slides, and mobile screens. Resolution reaches 2K via the API. Pricing is token-based: $0.006 per image at low quality and $0.211 at high, for a 1024x1024 output. A human designer, for reference, costs considerably more and requires breaks.
What happens next
Developers can access GPT Image 2 via API under the name gpt-image-2, at $8 per million input tokens and $30 per million output tokens. The model will improve. The benchmarks used to measure that improvement were, of course, designed by humans — who remain, for now, the ones deciding what counts as good.