Microsoft's MAI-Image-2.5 Climbs Image Arena, Joins Top 4 Labs
Microsoft AI's (@MicrosoftAI) in-house image model "MAI-Image-2.5" has entered the upper tier of the human-preference "Image Arena" rankings, lifting the company into the top four image-generation model labs. On the Design Arena leaderboard that prompted the announcement, it sits around Elo 1258 (in the #6–#7 band), level with Google's Gemini 3 Pro Image Preview (Nano Banana Pro).
Microsoft rolled out the model in stages from around May 26 to June 2, 2026. According to its official announcement, MAI-Image-2.5 is billed as "MAI's strongest image model yet," launching in two variants: the high-fidelity "MAI-Image-2.5" and the faster, lower-cost "MAI-Image-2.5-Flash." On Arena.ai's official benchmarks, it debuted at #3 with an Elo around 1254 on the Text-to-Image Arena, and rose to #2 with an Elo near 1401 on the Single-Image-Edit leaderboard, reportedly edging out the likes of Nano Banana 2 by about 10 points.
These rankings are produced by showing users images from multiple models on random prompts and having them blindly vote for the one they prefer, with results computed via Elo. That mechanism—judging by real user preference rather than marketing claims—is becoming a de facto industry standard. Microsoft, which had integrated third-party models like DALL·E into Copilot, has accelerated its own model development through the MAI team, progressing from MAI-Image-1 to MAI-Image-2 and now lifting quality with 2.5. The company highlights gains over the prior model in text rendering, commercial imagery, stylized illustration, prompt adherence, and visual reasoning around lighting, scale, and spatial relationships.
The model is available across Arena.ai, MAI Playground (playground.microsoft.ai), the developer-facing Azure AI Foundry, image generation in PowerPoint, and photo editing and background removal in OneDrive, as well as via OpenRouter. Pricing via Foundry runs about $47 per 1M image-output tokens for MAI-Image-2.5 and $19.50 for the Flash version, offering a lower-cost option.
Reaction spread quickly on X after the announcement, with the official Arena post drawing high engagement and CEO Mustafa Suleyman posting immediately. Developers and users praised it as "beating Nano Banana 2," "strong at editing," "good for text and commercial design," and "practical thanks to PowerPoint/OneDrive integration." At the same time, comparisons with GPT-Image-2, Grok Imagine, and Nano Banana 2 drew critical notes—"sometimes weaker on photorealism and specular reflections," "poor at celebrity likeness," and "fails on some prompts"—with a prevailing sense that "#3 is fair but not top-tier, though it holds up in editing." Both Microsoft and users flagged safety considerations, including potential bias and recommended review for sensitive use cases.