GPT Image 2

What it is

GPT Image 2 is an image generation model that OpenAI released on 21 April 2026.^[1] OpenAI also presented the same model under the name ChatGPT Images 2.0, linking it to its consumer chatbot product.^[2] In OpenAI's documentation it is described as a state-of-the-art model for fast, high-quality image generation and editing, with two further attributes singled out: flexible image sizes and high-fidelity image inputs.^[1]

That description points at what separates an image model from a text model. The output is a picture, so the useful knobs are things like canvas size and how faithfully an input image is preserved, rather than tokens and context windows.

How it is used

The model accepts text and images as input and returns images as output.^[1] Developers reach it through three endpoints — image generation, image editing, and Batch — and it supports inpainting, the operation in which a region of an existing picture is regenerated while the rest is left in place.^[1] The interface is otherwise narrow by design: audio, video, streaming, function calling, structured outputs and fine-tuning are all unsupported.^[1] In practice that makes it a picture-making component inside a larger application rather than a general conversational model.

OpenAI's documentation places the model on two of its own scales, rating its performance "higher" and its speed "medium".^[1] Access was not universal: the model was not offered to free-tier API accounts, and its snapshot alias was gpt-image-2-2026-04-21, the date embedded in the name matching the release day.^[1]

Where it came from

GPT Image 2 is the successor of GPT Image 1, released in 2025.^[4] The naming follows the pattern OpenAI uses for its other model families, where a number marks a generation and a date-stamped snapshot pins a specific build so that applications do not silently change behaviour underneath their users.

Its arrival also reorganised the older line. The release was followed by the deprecation of the earlier gpt-image-1-mini and gpt-image-1.5 models in June 2026, with shutdown set for 1 December 2026.^[4] Deprecation notices of this kind give developers a fixed window in which to migrate existing code to the newer model.

Where it sits in the line

GPT Image 2 did not stay the newest model for long. Two successor lines followed: GPT Image 2.5 Sunburst and Flare on 8 September 2026, and GPT Image 2.6 Sunburst and Flare on 4 November 2026.^[2][5] OpenAI describes the Sunburst models as aimed at detailed visual work and the Flare models at fast iteration.^[3]

The split is a common pattern in model families: one variant optimised for output quality on demanding tasks, another for speed and repeated trial-and-error. Read against GPT Image 2's own ratings — higher performance, medium speed — the later pair can be seen as taking the two ends of that single trade-off and offering them as separate choices.^[1][3]

For anyone following the field, GPT Image 2 is a useful marker of a particular moment: a single general image model, positioned as state of the art at launch, succeeded within months by a branching family, while the previous generation of smaller models was retired on a published schedule.^[1][4]