Back in October 2025, Microsoft announced MAI-Image-1, its first in-house image generation model to compete against the likes of OpenAI and Google. Although it was a contender that trailed from afar in this space, things changed when MAI-Image-2 was released last month. This model is currently occupying the third spot on the Arena leaderboard, and now, Microsoft is attempting to solidify its position as a viable option for consumers through a new variant with significant benefits.
MAI-Image-2-Efficient is the new spin-off from MAI-Image-2, which Microsoft already touts it as its best text-to-image model capable of reliably producing photorealistic images. As the name suggests, it prioritizes efficiency, which also brings down overall costs. In the Redmond tech firm"s testing, MAI-Image-2-Efficient was 22% faster and 4x more efficient than to MAI-Image-2. It"s also 40% faster on average than other hyperscaler models like Gemini 3.1 Flash (high reasoning), Gemini 3.1 Flash Image, and Gemini 3 Pro Image.
Microsoft"s latest image generation model is also 41% cheaper than the base MAI-Image-2 model. That said, both have their uses. MAI-Image-2 is ideal for generating complex, higher fidelity images with some textual detail. Meanwhile, MAI-Image-2-Efficient is designed to be more suitable for production in volume while optimizing cost. As such, it"s great for market creatives designing branding assets or making mockups. However, it is also capable of handling in-image text like labels and headlines.
The good thing is that MAI-Image-2-Efficient is available right now in Microsoft Foundry, which is a platform to build, optimize, and govern AI apps and agents. It can also be tested in MAI Playground, which is essentially an experimental lab where you can try out various MAI models as long as you sign in first. Microsoft has hinted that it"s working on more models, which should come as no surprise, so expect to see more announcements in this area soon.