GPT Image 2.5

OpenAI did not ship GPT Image 2.5 as a single model. It shipped two variants with identical controls and identical pricing that differ in where they spend their time. Flare is the one OpenAI recommends by default: it produces higher-quality output than GPT Image 2 while cutting latency by about half, which is the trade that matters when you are generating four candidates per brief, iterating on ad variations, or feeding a social calendar. Sunburst goes the other way. It takes longer, and it spends that time on intricate detail and on control across edits — the packaging mock with forty words of copy that all have to be right, the retouch that has to move one thing and disturb nothing else. On Genfire the two sit one chip apart in the same panel, so the sensible habit is to draft on Flare and switch to Sunburst for the final pass when the output has to hold up under a loupe.

The editing surface is where 2.5 earns its keep. Both tiers take up to 16 reference images in a single request, so a product, a model, a background, a logo, and a type reference can all inform one generation rather than being stitched together across several. Masked inpainting constrains a change to a region you paint, which is how you fix one element of an otherwise finished image without regenerating the whole frame. Quality is exposed as three tiers — low, medium, high — and billed with the same multipliers GPT Image 2 uses, so a low-quality draft round is cheap and only the approved direction gets rendered at high. Because GPT Image 2.5 costs the same as GPT Image 2 on Genfire, there is no financial reason to stay on the older model; Flare is a straight upgrade in both quality and speed, and Sunburst is an option GPT Image 2 never had.

What GPT Image 2 was good at carries forward: the prompt comprehension that gets counts, spatial relationships, and in-image text right, and the world knowledge that makes "a 1970s Polaroid" or "a Bauhaus poster" land without a paragraph of explanation. On Genfire that arrives without a ChatGPT subscription — both tiers run in the browser studio, on the REST API as image.gpt_image_25_flare and image.gpt_image_25_sunburst, through the MCP server, and from the CLI. Outputs flow downstream into upscaling, background removal, the ad studio, or a video model as a start frame, and the same brief can be A/B tested against Nano Banana, Seedream 4.5, or Muse Image on one credit balance. Genfire is pay-as-you-go: credit packs start at $19, purchased credits last 12 months, optional monthly plans start at $29, and every purchase includes a commercial license.