✳ the wire · analysis
OpenAI's own API spec names GPT Image 2.6 before any launch
An email when Next GPT Image (2.6 / 3) answers on a public API, about 15 minutes after members, plus every frontier launch · no card

OpenAI's published API specification now names an image model the company has not announced. A change pushed to OpenAI's openai-openapi repository at 20:56 UTC on October 28 adds a 'markup' option to the image edit endpoints, and its description says the option 'Supports GPT Image 2.5 and 2.6.'
WHAT THE SPEC SAYS
Markup is a transparent PNG laid over the first image to guide an edit. It can be passed as a file ID, an HTTPS URL or a base64 data URL, it has to match that image's dimensions after orientation and contain transparent pixels, and it cannot be combined with a mask.
WHAT IS NOT THERE YET
The spec's model lists still stop at GPT Image 2.5, in its Flare and Sunburst tiers and their September 8 snapshots. No 2.6 model id appears anywhere in the file as of the 04:05 UTC October 29 version we read, so this is a name in a description, not a model you can call.
THE CONTEXT
GPT Image 2.5 shipped on September 8 as two API tiers, Flare and Sunburst. On October 18, three 'aurora alpha' ids briefly appeared in OpenAI's API, per @scaling01, and @LuminaBench saw them generating images in Arena. Nothing ties them to 2.6.
WHAT WE DO NOT KNOW
When GPT Image 2.6 ships, what it costs and what changes from 2.5. OpenAI has said nothing about it. The spec line was first spotted on X by @Mr_Salio.
Source: OpenAI's openai-openapi repository on GitHub: commit aa880040 'Add markup property to image editing endpoints' (Oct 28 20:56 UTC) and the current openapi.yaml (both read directly) ↗ · Next GPT Image (2.6 / 3) tracker · the bench index

