Meta just crashed the AI image generation party, and it brought receipts. The company’s new Muse Image model has landed the number two spot on Arena.ai’s Text-to-Image Arena leaderboard, posting an Elo score of 1280 based on 7,715 community votes. Only OpenAI’s GPT Image 2, sitting comfortably at 1385, ranks higher.
What Muse Image actually does
Announced on July 7, 2026, by Meta Superintelligence Labs, the Muse Image model features agentic tool use, meaning it can actively leverage web search and coding functionalities during the image creation process. The model also exhibits what Meta calls emergent self-refinement during generation, using test-time compute scaling to improve results before you ever see the final image.
Muse Image also secured the number two position in both the single-image edit and multi-image edit arenas on the same platform. Its companion model, Muse Video, grabbed the number three spot in Arena.ai’s Text-to-Video rankings.
Where you can actually use it
