GPT Image 2

GPT Image 2 is OpenAI's current image generation and editing model, released as the snapshot gpt-image-2-2026-04-21. It replaces the whole earlier line — DALL·E 2 and 3, gpt-image-1 and its mini and 1.5 variants — which OpenAI is retiring in favour of this single model. Its most practical departure from its predecessors is sizing. Where gpt-image-1 offered three fixed shapes, GPT Image 2 accepts any resolution that fits a set of constraints: no edge longer than 3840 px, both edges multiples of 16, an aspect ratio no more extreme than 3:1, and a total between roughly 0.65 and 8.3 megapixels. In practice that covers everything from a square 1024 px thumbnail to a 4K poster without cropping or upscaling afterwards. The model takes text and images as input and returns images. Reference pictures are always processed at high fidelity — unlike the previous generation, there is no fidelity setting to get wrong. Quality is chosen as low, medium, high or auto, and generation can be streamed with up to three partial previews so a user sees the picture forming. One limitation carried over from no predecessor: transparent backgrounds, supported by gpt-image-1, are not available here. Billing is by token rather than by picture: USD 8 per million input tokens, USD 2 for cached input and USD 30 per million output tokens, halved in the Batch API. A high-quality 1024×1024 image works out at roughly USD 0.21. It is reachable through the image generation and edit endpoints and through Batch.

▸News

Six image models become one: OpenAI's deprecation list dates the end of every generator it ever sold

8/14/2026 · OpenAI API deprecations

▸Videos

No videos yet.

▸Reviews