Microsoft on March 19, 2026, introduced MAI-Image-2, the second text-to-image model built entirely by its in-house Microsoft AI group. The company says the model debuted in the top three of the Arena text-to-image leaderboard, a crowdsourced ranking in which users compare outputs from different models side by side, placing it just behind models from Google and OpenAI.
Microsoft says MAI-Image-2 is being integrated into Copilot and Bing Image Creator, which means it will reach a large consumer audience through products that previously relied heavily on OpenAI’s DALL-E and GPT image models.
A sharp climb from the first version
The first model in the series, MAI-Image-1, was released in October 2025 and entered the same leaderboard in the top ten rather than the top three. The jump between versions in roughly five months suggests Microsoft has moved image generation from an experimental project to a priority, according to coverage of the launch in The Next Web and Neowin.
In its announcement, Microsoft emphasized photorealism, accurate rendering of text inside images and faithfulness to detailed prompts, areas where earlier consumer image generators have often fallen short. Those claims come from the company; independent evaluations beyond the Arena ranking were not available at launch.
Part of a push to rely less on OpenAI
MAI-Image-2 comes from the team led by Mustafa Suleyman, the DeepMind co-founder who joined Microsoft in 2024 to run its consumer AI business. His group has been building a family of first-party models under the MAI label, including a speech model and a text model, alongside Microsoft’s long-standing partnership with OpenAI.
The strategy is widely read as an effort to give Microsoft more control over the models that power its own products and to reduce the cost and dependency that come with licensing another company’s technology. Microsoft and OpenAI restructured their agreement in 2025, and Microsoft has since said publicly that it intends to develop frontier models of its own.
What it means for the market
Image generation has become one of the most crowded corners of the AI market, with Google’s Imagen and Nano Banana models, OpenAI’s GPT image tools, Black Forest Labs’ Flux and several Chinese labs all competing on quality and speed. Microsoft’s entry into the top tier gives the company a credible alternative to ship across Windows, Microsoft 365 and Bing without depending on a partner.
Microsoft has not published pricing for API access to MAI-Image-2 or said whether it will be offered through Azure to outside developers. For now, the model’s main distribution channel is Microsoft’s own consumer products.
