Microsoft's MAI Image 2.6 debuts second on Arena

Microsoft AI has announced MAI Image 2.6, a point release of its in-house text-to-image model. It currently sits second on the Arena text-to-image leaderboard.

According to the company, 2.6 gains 79 Elo overall compared with MAI Image 2.5 and improves in every Arena text-to-image category. Text rendering shows the biggest jump, at 91 Elo. Microsoft says the result puts MAI Image ahead of leading models from Meta, Google and xAI. The announcement does not name the model in first place or list the competing models it beat.

Microsoft lists three areas of improvement: text rendering, portraits and 3D imagery, and commercial and photorealistic output for product, branding and cinematic work. The company presents the release as the next step in the series. In its telling, MAI Image 2 improved photorealism and text, and 2.5 moved into professional-grade imagery and editing.

Microsoft also mentioned support for multiple reference images, richer grounding, and more control over reasoning, format and resolution. It gave no details and said it will "share more on that soon."

For now, MAI Image 2.6 is available for text-to-image generation only on Arena. Microsoft says it will reach MAI Playground later this week, followed by Microsoft Foundry and other products. The company did not announce pricing, API terms or technical details such as model size.

All of the performance claims come from Microsoft's own announcement. No independent evaluation is cited.