Microsoft AI has unveiled MAI-Image-1, its first in-house image generator, which has entered the LMArena Text-to-Image leaderboard in the top ten. The model is framed as creator-focused, avoiding generic style patterns, and has been evaluated with feedback from creative professionals. It is positioned for future integration into Microsoft products.
As of October 12, 2025, LMArena lists MAI-Image-1 at rank No. 9 with a preliminary score of 1096, alongside models from Google, OpenAI, and Bytedance. The leaderboard displays community vote totals and confidence intervals, indicating an early but competitive standing.
Meet our third @MicrosoftAI model: MAI-Image-1
— Mustafa Suleyman (@mustafasuleyman) October 13, 2025
#9 on LMArena, striking an impressive balance of generation speed and quality
Excited to keep refining + climbing the leaderboard from here!
We're just getting started.https://t.co/33BiNfIjPg pic.twitter.com/FMaXqiVIvS
The model targets photorealism, adeptly handling lighting, reflections, and landscapes, and claims faster iteration than larger, slower peers. Microsoft cites rigorous data selection and nuanced evaluation to enhance visual diversity and practical utility for creators. No public API has been announced, with availability implied to begin through Microsoft products rather than direct access.
Microsoft AI’s lab recently previewed MAI-1-preview, a mixture-of-experts model trained on roughly 15,000 NVIDIA H100 GPUs, and shipped MAI-Voice-1 in Copilot. The group notes that a next-gen GB200 cluster is now operational, signaling ramped compute for upcoming releases.
