Benchmarks · Arena Elo · July 5–7, 2026

Muse Video Benchmarks — Arena Elo Rankings

How Muse Video ranks: No. 3 on Arena for text-to-video by human-preference Elo, with the search, self-refinement, and test-time-compute gains that lift its Muse Image twin. Figures time-stamped July 5–7, 2026.

Arena No. 3 · text-to-videoArena No. 2 · image editingElo & ablations

Where Muse Video ranks

On Arena, the human-preference leaderboard, Muse Video ranks No. 3 for text-to-video by Elo as of July 5, 2026 — behind Google's Gemini Omni Flash and Bytedance's Seedance 2.0, and ahead of the rest of the field. Its Muse Image twin ranks No. 2 for text-to-image and image editing over the same window. The charts and leaderboard below reproduce Meta's published results; all figures are time-stamped July 5–7, 2026.

This is an independent, unofficial page and is not affiliated with Meta. The ablation figures (search, self-refinement, test-time compute) describe the Muse Image twin that shares Muse Video's pretraining base.

Muse Video benchmarks & public signals

Muse Video ranks No. 3 on Arena for text-to-video by human-preference Elo, while its Muse Image twin improves measurably with search, self-refinement, and test-time compute.

Improvement with search

With searchWithout search
Identities
70.2%
29.8%
Brands
67.9%
32.1%
Landmarks
67.3%
32.7%
Facts
56.6%
43.4%
50%
Muse Image improves with search tool use. Win rate from internal ablation.

Improvement with self-refinement

With self-refinementWithout
Text-to-Image
57.1%
42.9%
Single-Image Edit
56.3%
43.7%
Multi-Image Edit
56.6%
43.4%
50%
Muse Image improves with emergent self-refinement. Win rate from internal ablation.

Improvement with test-time compute

Reasoning (with tools)Reasoning (no tools)Best-of-N
10201000980960Test-Time Compute (Log Scale)Elo
Muse Image improves with scaling test-time compute. Elo from internal ablation.

Text-to-Video Arena

#Lab and ModelScore
1Gemini Omni FlashGoogle1527 ±13
2Seedance 2.0Bytedance1482 ±10
3Muse VideoMeta1459 ±15
4Happyhorse 1.0Alibaba-ATH1430 ±13
5Sora 2 ProOpenAI1366 ±8
6Veo 3.1 Audio (1080p)Google1364 ±11
7Veo 3.1 AudioGoogle1364 ±14
8Veo 3.1 Fast AudioGoogle1362 ±11
9Veo 3.1 Fast Audio (1080p)Google1360 ±10
10Grok Imagine Video (720p)xAI1352 ±8
Source: Arena AI Leaderboard · July 5, 2026
Known Muse Video gaps: audio-video synchronization and physically accurate fast motion.

Figures as of July 5–7, 2026, from Meta's published results.

Known performance gaps

Meta is candid about where Muse Video still falls short. The two named gaps are audio-video synchronization — keeping the native audio perfectly locked to fast on-screen action — and physically accurate fast motion, where quick, complex movement can still break realism. Publishing these limitations alongside the Arena rank is part of why the Muse Video benchmarks read as credible rather than cherry-picked.

How to read these figures

The Arena table is a human-preference Elo, so a higher score means people picked that model's text-to-video output more often in head-to-head comparisons. The bar charts are internal ablations showing win rate with a capability versus without it — search grounding and self-refinement each move the needle. The line chart shows Elo rising as test-time compute scales. For the plain-English background, see what Muse Video is, or read about the Muse Image twin these ablations measure.

Muse Video benchmarks FAQ

Muse Video ranks No. 3 in human-preference Elo for text-to-video on Arena as of July 5, 2026, behind Gemini Omni Flash and Seedance 2.0. Its Muse Image twin ranks No. 2 for text-to-image and image editing.

Meta names two gaps: audio-video synchronization and physically accurate fast motion. These are published alongside the benchmark results.

All Muse Video and Muse Image benchmark figures on this page are time-stamped as of July 5–7, 2026, from Meta's published results.

See what Muse Video can do

Generate a real AI visual and preview the native-audio motion — free, no account, no credits.

Try Free