OpenAI: GPT Image 2.5 Sunburst

GPT Image 2.5 Sunburst is an image generation and editing model from OpenAI. It sits within the GPT Image 2.5 series and is positioned as the precision-oriented tier of that family, aimed at detailed creative work. On api14 the model id is openai/gpt-image-2.5-sunburst, which is the identifier you use when sending requests through the gateway.

The model accepts text and images as input and produces images as output, so it handles both generation from a prompt and editing or transformation work that starts from reference images. It supports up to 16 input references, which lets you guide a result with a set of supplied images rather than a single one. The context window is 400,000 tokens, with a maximum output of 360,000. Supported request parameters are max_tokens, response_format, seed and structured_outputs. Seed is useful when you need repeatable results across calls, and structured_outputs lets you constrain the shape of returned data.

Because it takes both text and images as input and returns images, the practical uses follow from that combination. You can generate original images from written descriptions, restyle or modify existing images while keeping the parts you want, and composite ideas drawn from several reference images at once. The ability to supply many references makes it a reasonable fit for tasks where a consistent subject, product or visual style has to carry across a set of outputs. Because it is described as the precision-oriented tier, it is the more appropriate choice when fine detail in the output matters more than speed.

Calling it through api14 works like any other OpenAI-compatible endpoint. Send a request to the chat completions route and set the model field to openai/gpt-image-2.5-sunburst. Alongside the standard parameters, the image controls include aspect_ratio, which accepts 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9 or auto; quality, which accepts auto, low, medium, high, xhigh or max; and background, which accepts auto, transparent or opaque. You can also set n between 1 and 10 to request several outputs in one call, input_references between 0 and 16, and output_compression between 0 and 100.

Note that the model outputs images only, so it is not the right pick if you need text returned from the same call or a purely conversational response. For text work, another model in the catalogue will fit better. Within image tasks, if you do not need the precision tier, a lighter model may be sufficient.