Microsoft's MAI-Image-2 cracks the top 3, challenging Google and OpenAI

With MAI-Image-2, Microsoft has successfully transitioned from a trailing contender to a top-tier powerhouse in the generative AI space.

Back in October 2025, Microsoft announced its in-house image generation model, MAI-Image-1, to compete against OpenAI's gpt-image-1 and Google's Gemini/Imagen models. Later, Microsoft expanded the availability of this model to Bing.com/create and Bing mobile app users.

When the model was announced in Arena's Image Arena, this new model debuted at the #9 position, significantly lagging behind other AI labs. Microsoft claimed that they trained the model to prevent it from producing repetitive or overly stylized outputs and also that it is significantly faster compared to other large image generation models on the market.

Today, the Microsoft AI team introduced MAI-Image-2, its second-generation image generation model, featuring significant improvements. MAI-Image-2 can now generate images with better natural light, accurate skin tones, and more natural environments, enabling the creative community to spend less time in post-production.

Microsoft has also significantly improved the text rendering capability of the model to make it more suitable for the creation of infographics, slides, diagrams, and more. Microsoft also claims that MAI-Image-2 is now better at generating cinematic and hyper-detailed images involving surreal concepts, ornate compositions, and ambitious worlds.

Our new image generator MAI-Image-2 is out! Available now on MAI Playground for everything from lifelike realism to detailed infographics.

Our team has been pushing immensely hard for this release, and we are now among the top models out there: #3 family on @arena.

Check out… pic.twitter.com/2FkPUyrGYC
— Mustafa Suleyman (@mustafasuleyman) March 19, 2026

In the Arena leaderboard for image generation models, MAI-Image-2 is now ranked #3, trailing only Google gemini-3.1-flash-image-preview and OpenAI's gpt-image-1.5-high-fidelity.

Copilot and Bing Image Creator users can use the MAI-Image-2 model to generate images once the rollout is complete. Users who want to use the model straight away can visit MAI Playground and give it a try. Developers interested in using MAI-Image-2 inside their apps and services can access it via API on Microsoft Foundry.