MAI-Image-2.6 debuts at No.2 in Text-to-Image Arena
Microsoft AI's newest image model entered Arena's text-to-image leaderboard at second with 1336 points, up from 2.5's tenth place and 45 behind GPT Image 2.
Arena said on Monday that MAI-Image-2.6, the latest text-to-image model from Microsoft AI, entered its Text-to-Image leaderboard in second place with 1336 points. Its predecessor, MAI-Image-2.5, sits at tenth.
The numbers
The gap to the top is now 45 points: GPT Image 2 (Medium) holds first. Arena reported gains across every scored category, with the sharpest movement in 3D imaging and modelling, where the model went from sixth to first, alongside improvement in cartoon and illustration work. Arena has opened the model for public voting in Image Arena so the ranking can be tested against fresh human preference data rather than a launch snapshot.
An eight-place jump between consecutive versions of the same family is unusual at the top of this board. Microsoft's in-house imaging effort entered at ninth with MAI-Image-1 in October 2025, reached third with MAI-Image-2 in March 2026, then stalled — 2.5 debuted at third in May and slid as competitors shipped. Second place is the highest the family has ranked.
Context
Human-preference leaderboards are noisy and reward style as much as fidelity, and early Elo scores typically drift as vote volume accumulates. But the standing at the top of text-to-image has been stable for months, dominated by OpenAI's GPT Image 2 and Google's Imagen line, with Microsoft, Black Forest Labs and Midjourney trading positions below. A 45-point deficit is close enough that the next release cycle could change the order.
Why it matters
Microsoft distributes image generation through Copilot, Bing Image Creator and Azure API access rather than a standalone creative app, so a model at this quality level reaches an unusually broad default audience without users choosing it. The company also runs the largest commercial distribution channel for OpenAI's models — narrowing the gap with GPT Image 2 using its own weights strengthens Microsoft's hand in exactly the place that dependency has been most visible.