// news · multimodal · research-papers2026-05-25source: bytedance / techcrunch / oimi.ai

Seedance 2.0 holds Artificial Analysis Video Arena top with Elo 1,269 text-to-video, 1,351 image-to-video — ahead of Veo 3, Sora 2, Kling 3.0, Runway Gen-4.5

ByteDance Seedance 2.0 (released February 2026) continues to lead the independent Artificial Analysis Video Arena with Elo 1,269 in text-to-video and 1,351 in image-to-video — both ahead of every Western competitor measured. The standing has held for three months despite Veo 3.1, Sora 2 API release, Kling 3.0, and Runway Gen-4.5 all shipping into the same evaluation pool.

The persistence of the lead is what makes it newsworthy this cycle. Single-benchmark wins are common; three months of arena dominance against four well-funded competitors is a durable position. The Elo gap between Seedance 2.0 and the next-best Western model is large enough that a single new release from OpenAI or Google would need to be a clear capability jump (not an incremental update) to close it. Veo 3.1 closed some of the gap but didn't surpass; Gemini Omni Flash trails Seedance on raw generation quality per early leaks.

The geographic interpretation is what matters for the policy framing. Export controls on advanced GPUs to China were designed to slow Chinese frontier-AI capability development. Seedance 2.0's persistence at the top of video-generation benchmarks demonstrates the controls are not working for this product category. ByteDance trained Seedance with whatever chip mix was available to them in 2025; the result is benchmarking ahead of OpenAI, Google, Runway, and Kling. The policy question this raises — whether tighter controls would close the gap, whether different metrics would tell a different story, or whether the controls are simply mismatched to the engineering surfaces that determine video-generation quality — is the conversation the next US administration will have to engage on.

See our analysis →

Medium / Analyst Uttam — Gemini Omni Video Model vs Seedance 2 → · Oimi AI — Gemini Omni Flash Release and Seedance 2 Comparison → · Vo3 AI — Gemini Omni Google Unified Multimodal Video Model →