OpenArt Arena Launches Expert Blind-Test Leaderboard for Creative AI Models
On September 16, OpenArt launched Arena, a global ranking of AI image and video models focused on "creative intelligence." Evaluation is done via blind testing by creative industry professionals: model names and logos are hidden, judges vote between two outputs, and rankings are broken down by creators' actual work scenarios (film, advertising, animation, motion design, video editing, etc.).
Confirmed
- Arena is built by OpenArt, uses an expert blind-testing mechanism, and ranks image and video generation models by creative task type.
- Top three on the overall video leaderboard: ByteDance's Seedance 2.5 (11125 points), Alibaba's Wan 3.0 (11047 points), and ByteDance's Seedance 2.0 (1104x points — the number was truncated in the post).
- Rankings on sub-leaderboards such as advertising and film/TV differ markedly from the overall video board, with different models leading on different tasks.
Not yet confirmed
- MayorKingAI anticipated the rankings would spark controversy and asked followers to weigh in; the leaderboard's credibility still awaits validation from the creator community.
Why it matters
- azedai, after testing many models hands-on, argued that "which AI model is strongest" is the wrong question — the right one is "strongest at what task" — making general-purpose leaderboards of limited value; Arena's task-specific boards can give creators a starting point for model selection.
- Driven by expert blind testing rather than automated benchmarks, the leaderboard is closer to real creative production needs and could shape model selection habits in the creative industry.
2026-09-16 ~ 2026-09-16 · 6 related posts
Primary sources
- OpenArt Arena launches a pro blind-test leaderboard for creative AI models — azed_ai · 2026-09-16
- [source] Creative AI leaderboard: Seedance 2.5 tops video, Wan 3.0 close behind — minchoi · 2026-09-16
- OpenArt Arena's current top AI models ranked, sparking debate — azed_ai · 2026-09-16
- [source] OpenArt Arena ranks image and video models per task type, and the winners change by category — azed_ai · 2026-09-16