Microsoft is coming for the king of the AI image generator game, and on at least one metric, it didn’t miss. The company on Tuesday dropped two new text-to-image models, named MAI-Image-2.5, as part of a series of next-generation AI models. But how does it compare to the gold standard in image generation, Google’s Nano Banana?
When you think about Microsoft and its AI offerings like Copilot, you probably don’t think about creative work. The capabilities of these two new models — the regular 2.5 and a quicker Flash one — are designed to change your mind.
Read more: Microsoft Build: All the Biggest Copilot AI, Agent and Chip News From San Francisco
“They give you precise editing with incredible control and consistency,” Microsoft AI CEO Mustafa Suleyman said during the Build keynote Tuesday. “Flash is here for super-efficient production workloads, while 2.5 gives you that maximum fidelity and professional-grade performance.”
Precise editing of AI images is important for professional or enterprise work. This is an example Microsoft shared of how its new model can adjust images.
These are two of seven new AI models that Microsoft unveiled at its Build developer conference on Tuesday. It also released its first “reasoning” model, named MAI-Thinking-1. (A reasoning model is designed to spend more time and do more iterations on a response, ideally to generate a better one or solve a more complex question.) New generations of its voice and transcription models were also introduced, along with a coding model optimized to run on GitHub, which is also owned by Microsoft. We also got an in-depth look at how Microsoft sees the future of computing, and — spoiler — it’s all about agentic AI.
The new image model is available now in PowerPoint and in its enterprise marketplace, Foundry, and is rolling out now in OneDrive.
This is how Microsoft’s newest AI image models stack up against the competition.
Which is better, MAI-Image-2.5 or Nano Banana 2?
When it comes to AI image work, Google’s Nano Banana has dominated the creative AI space since its 2025 launch. It’s for good reason — its capabilities are industry-leading, even as it powers a deluge of slop and deepfakes. That’s why it was surprising to hear Microsoft announce that its new image model had beat out Nano Banana in one key way.
Microsoft’s MAI-Image-2.5 is better at image editing than Nano Banana 2, according to benchmarking shown on the Arena AI leaderboard, a popular measuring stick for AI models. But Microsoft is only getting the silver medal — OpenAI’s GPT-Image-2 currently holds the top spot.
The usefulness of an AI model isn’t solely defined by any one benchmark, however. Accessibility plays just as big a role. One simple question will determine which will be more useful for you: Do you use PowerPoint or Google Slides? That ease of access to Microsoft and Google’s respective AI models is going to influence which one you end up using. If you’re using your AI images for commercial use, you’ll want to make sure you have the rights, which may depend on whether you’re using the AI on an enterprise or individual plan.
