Microsoft recently released its latest generation text generation image model MAI-Image-2.5, and it immediately took the third place in Arena's latest text generation rankings, showing that its overall strength in the field of AI image generation is clearly catching up with industry leaders. Currently on this list, OpenAI’s gpt-image-2 ranks first with a score of 1388.
Microsoft said that MAI-Image-2.5 has comprehensively improved its performance in a variety of image styles, can better fit user prompts to generate images, and has significant improvements in text rendering and screen details.
Microsoft initially launched its self-developed image generation model MAI-Image-1 last year. The model ranked only 9th in Arena's Image Arena rankings, which was significantly different from the top models of other laboratories at the time. Microsoft then integrated the model into Bing.com/create and the Bing mobile application and made it available to users. In March this year, the Microsoft AI team released the second generation MAI-Image-2, which brought significant improvements in natural light performance, skin color restoration, etc., and broke into the top three of the Arena rankings for the first time, trailing only Google's gemini-3.1-flash-image-preview and OpenAI's gpt-image-1.5-high-fidelity. It is also simultaneously connected to Copilot, Bing Image Creator and Microsoft Foundry API for developers to call.
On this basis, the new generation MAI-Image-2.5 further strengthens the overall capabilities of the model. Microsoft said that MAI-Image-2.5 can more accurately understand and execute the user's text prompts, the generated images are more coherent in details and overall structure, and the rendering of text content is more reliable. The model's visual reasoning capabilities have also been enhanced, with a more accurate understanding of objects, lighting, proportions, scene structures, and spatial relationships, helping to generate images that are more in line with real scenes and user expectations.
Microsoft specifically emphasized that MAI-Image-2.5 has made the most obvious progress in text rendering, stylized illustrations and commercial images, including posters, packaging proofing, brand creativity and product images, etc., which can produce higher quality output. Compared with the previous generation model, the text in the images generated by the new model is clearer and sharper, the layout is more stable, and the brand visual elements are more prominent and have a more overall texture, which is more in line with the needs of commercial design and marketing applications.
Mustafa Suleyman, head of Microsoft AI, said on social platforms that MAI-Image-2.5 ranked third in the Arena Vincentian chart rankings, marking another important leap in the quality of image generation. He also predicted that at the upcoming Build conference, the Microsoft AI team will bring more related updates and releases. He said that this is just another step in Microsoft's efforts in the field of AI imaging, and it will continue to promote the iteration of model capabilities and product forms in the future.
Currently, MAI-Image-2.5 is open to all users for trial use through Arena, and users can directly compare and experience it with other models on the platform. According to Microsoft's plan, this new model will also be launched on MAI Playground and Microsoft Foundry in the next two weeks, providing ordinary users and developers with more diverse generative image creation and integration options.