Meta Muse Completes Its Arena Presence: Image and Video Models Enter Leaderboards 60 Days After Language Debut
Meta’s Muse models now cover every Arena category.
Arena added muse-image to the Text-to-Image and Image Edit leaderboard on July 7, and muse-video to the Text-to-Video leaderboard on the same day. The language model Muse Spark debuted two months ago at fourth on Artificial Analysis’s Intelligence Index. With the generative media additions, Meta’s Muse suite now competes across all three major Arena evaluation categories: text, image, and video.
Elo scores for muse-image and muse-video will accumulate over the coming days as battles populate the leaderboards. Initial ratings are not yet available.
The Video Fight
ByteDance’s SeedDance 2.0 holds both top spots in Arena’s video leaderboard at Elo 1450, 79 points ahead of Google Veo 3.1 in second. SeedDance 2.5 then extended native generation to 30 seconds and raised the reference image ceiling to 50 slots — three capability jumps in the same release cycle. The lead has been structurally stable since late Q1 2026.
muse-video enters that category without published frame quality metrics, motion coherence scores, or direct SeedDance comparison data. Meta’s physical AI infrastructure — MTIA 400 inference silicon and the Muse Spark language pipeline — suggest significant investment in the underlying generation stack, but Arena battles will establish the competitive position.
The Image Fight
Microsoft’s MAI-Image-2.5 entered Arena’s text-to-image leaderboard in June at Elo 1,254, taking third place. Luma’s uni-1.1-max holds the top human preference ranking at $0.10 per image. HiDream O1-Image and Google Imagen 4 are the other primary competitors.
krea-2-turbo also joined the image Arena on July 2, establishing Krea’s turbo variant alongside the full model the lab debuted at no. 6 on Artificial Analysis earlier this year. muse-image enters a category that has seen four significant new entrants in the past 30 days.
Why the Full Suite Matters
Meta is now the only major AI lab competing simultaneously in Arena’s text, image, and video evaluation tracks. OpenAI’s GPT Image 2 is available via API but has not accumulated a leaderboard Elo ranking on Arena. Anthropic has no image or video generation entry.
The Muse naming convention also signals a unified product identity across modalities — a branding strategy that positions Meta against the vertically integrated output of Google (Gemini text, Imagen, Veo) and ByteDance (Seed text, SeedDance video). Whether the scores support the positioning will emerge within the week.