Microsoft's MAI-Image-2.5 Hits #2 Globally, Challenging OpenAI's Image Dominance

Microsoft just posted one of the strongest image model debuts in recent memory. MAI-Image-2.5, the latest release from the company's in-house MAI Superintelligence team, ranks #1 for GPT Image 2 with an Elo of 1339, while MAI-Image-2.5 lands at #2 with an Elo of 1276 on the Artificial Analysis Text-to-Image Arena. On the editing side, the image editing leaderboard puts GPT Image 1.5 at #1 (Elo 1261), GPT Image 2 at #2 (Elo 1260), and MAI-Image-2.5 at #3 (Elo 1251). For a model family that didn't exist eight months ago, that's a remarkable trajectory.

What actually shipped

Microsoft is launching MAI-Image-2.5 for maximum fidelity, and MAI-Image-2.5-Flash for fast, scalable production workloads. Both handle text-to-image generation and image editing in a single model, supporting up to roughly 1MP output at flexible aspect ratios with a 32K token context window.

The headline capability improvements over the previous MAI-Image-2 are significant:

• MAI-Image-2.5 delivers an overall +75 point improvement over MAI-Image-2, with the largest gains in Text Rendering (+107) and Cartoon, Anime & Fantasy (+90).
• The model understands scene structure, lighting, scale, and spatial relationships, helping it make edits that fit the image context, such as adding an object with the right perspective and shadows.
• MAI-Image-2.5 supports precise, localized edits -- from replacing an object or updating text to removing motion blur -- without changing the rest of the image.
• MAI-Image-2.5 preserves facial identity across edits, maintaining recognizable likeness even through changes in pose, expression or viewpoint.

Microsoft positions 2.5 as a step change over MAI-Image-2, with the largest gains concentrated in text rendering, stylized illustration, and commercial imagery. That last point matters more than it sounds: legible text inside generated images -- think posters, packaging, infographics -- has historically been one of the hardest problems for diffusion-based models.