Microsoft's MAI-Image-2.5 ranks #2 in Image Edit Arena
Microsoft AI's image model "MAI-Image-2.5" has landed at #2 in the "Image Edit Arena (Single-Image Edit)" with a score of 1401, behind only OpenAI's "gpt-image-2 (medium)." Announced by Arena.ai, which runs the evaluation, on June 2, 2026, the company said that "at its price tier, no model scores higher on Arena," noting that it advances the "Pareto frontier" representing the optimal balance of performance and cost (Arena official post).
According to Arena.ai, MAI-Image-2.5 scored 1401 (±8, preliminary) over 5,625 matchups in the single-image edit category, pulling more than 10 points ahead of Google's "Nano Banana 2," xAI's "Grok Imagine," and OpenAI's "ChatGPT-Image-Latest-High Fidelity" (Arena detail post, Image Edit leaderboard). The top spot belongs to "gpt-image-2 (medium)" at 1466, leaving only a narrow gap to the leader.
The model was originally unveiled on May 26, 2026, as a text-to-image model, debuting at #3 in the Text-to-Image Arena with a score of 1254 (±8). That marked a +72-point improvement over the previous MAI-Image-2, drawing attention as a sign of rapid progress (Morphic, Microsoft AI official). Following its text-to-image showing, it has now broken into the upper ranks in image editing as well.
The MAI-Image series began with "MAI-Image-1," which debuted at #9 in LMArena's text-to-image in October 2025, followed by "MAI-Image-2" in March 2026 and "MAI-Image-2.5" in May 2026, climbing the rankings in a short span. Development is led by Microsoft's internal "MAI Superintelligence Team," and the model is positioned as a practical, work-oriented tool reflecting feedback from creators such as photographers and designers (Medium). Its strengths are cited as reliable text rendering, instruction following, stylized illustration, commercial uses such as product photography and brand visuals, and visual reasoning involving lighting, depth, and spatial relationships (Morphic).
On the competitive front, The Decoder reported that MAI-Image-2.5 pulls even with Google's Nano Banana 2 on benchmarks and sits behind OpenAI's Image-series models, emphasizing professional uses such as product photography and brand design (The Decoder). Gigazine also introduced it as a world-class image generation AI with strengths in text rendering (Gigazine). In terms of availability, the model can be used immediately on Arena and is slated to roll out to MAI Playground and Microsoft Foundry within one to two weeks, with future integration into PowerPoint, OneDrive, and Copilot also mentioned.
Reaction on X has been largely positive. Arena's official account and the Microsoft AI team actively shared the news, and a Microsoft MVP account pointed to its practical integration, noting it ranked #2 on Arena for image editing, ahead of Nano Banana 2, is live in PowerPoint, and is rolling into OneDrive (FlowAltDelete). Others praised specific improvements such as "a +72-point jump is a real signal" and "text rendering, composition, and commercial image quality" (Wes Roth). Within the range observed this time, notable criticism was scarce, but as it remains in preview, hands-on verification including specific use cases and failure examples is expected to grow.