Microsoft launches MAI-Image-2
Microsoft rolled out MAI-Image-2 — now ranked #5 on text‑to‑image leaderboards — along with an MAI Playground for model experimentation, signaling a renewed push into creative AI tooling. The releases give developers new options for integrating image generation into apps and prototypes. (x.com)
Microsoft’s announcement was posted March 19, 2026 and named WPP as a select enterprise getting API access today while promising broader developer availability on Microsoft Foundry soon (microsoft.ai). (microsoft.ai) Arena.ai’s text‑to‑image leaderboard currently lists MAI‑Image‑2 at rank 3 with a score of 1189±8, behind Google’s gemini‑3.1‑flash‑image‑preview and OpenAI’s gpt‑image‑1.5‑high‑fidelity. (arena.ai). (arena.ai) Early reviews highlight photorealism, improved skin‑tone fidelity, and stronger in‑image text rendering as MAI‑Image‑2’s technical wins compared with Microsoft’s first generation, based on hands‑on tests reported by Decrypt. (decrypt.co). (decrypt.co) Coverage of the public preview documents conservative usage limits: outlets report strict content filters and 1:1‑only outputs, and third‑party summaries have noted a 30‑second cooldown and a reported daily cap of 15 images in early testing. (decrypt.co) (binance.com). (decrypt.co) Microsoft specified enterprise API access is live for partners like WPP and reiterated that the model will be exposed to developers via Microsoft Foundry, where Foundry playgrounds are documented as prototyping sandboxes without production SLAs. (microsoft.ai) (learn.microsoft.com). (microsoft.ai) The announcement comes from Microsoft’s AI Superintelligence team after a leadership reshuffle that put Mustafa Suleyman more directly in charge of frontier model efforts, a personnel move noted in contemporaneous reporting. (thenextweb.com). (thenextweb.com)
