A new image generation model has arrived on the Arena leaderboard. Scoring a 2nd place rank, MAI Image 2.6 brings a 79 Elo increase over its predecessor along with major upgrades to text rendering. The model is currently available on Arena with a roll out to wider consumer platforms scheduled for later this week.
The developers are actively climbing the competitive ranks. This update positions the software ahead of rival generators built by Meta, Google, and xAI. The most obvious gains show up in the text rendering category, which jumped by 91 Elo points. Generating readable English inside an image has historically been a massive struggle for artificial intelligence, so this specific bump is a major win.
This release targets some of the most common complaints from digital creators. Portraits look cleaner and 3D imagery displays much better depth. Commercial designers will also find that branding projects, product mockups, and cinematic outputs look much closer to actual photography. The developers accomplished this by systematically addressing quality issues across every evaluation category.
Future updates will expand the toolset even further. The team plans to add support for multiple image references, cleaner grounding, and precise control over formatting and final resolution. While those updates are still in the pipeline, you can test the core generator right now on the Arena platform. It will arrive on the official Playground site in a few days and will hit Microsoft Foundry shortly after.
