Wan 2.1

Wan 2.1 is Alibaba's Tongyi Wanxiang video foundation model, open-sourced in February 2025 in four checkpoints: T2V-14B, T2V-1.3B, I2V-14B-720P, and I2V-14B-480P, covering text-to-video and image-to-video generation. The 14B flagship targets cinematic-quality output with strong complex-motion handling; the 1.3B variant runs on consumer GPUs (reportedly ~8.2GB VRAM for 480p output). It was the first video generation model to support text effects in both Chinese and English.

At launch, Alibaba reported a VBench overall score of 86.22% (provider-reported), leading in dynamic degree, spatial relationships, color, and multi-object interactions — claimed as the only open-source model in the VBench top five. This figure has been repeated across launch coverage but I found no independent re-run of the VBench numbers, so treat it as a provider claim.

Benchmarks (all provider-reported, from Alibaba's launch materials):

• VBench overall: 86.22% — tops the leaderboard at release (provider-reported; independent verification not found)
• VBench dimensions: leads in dynamic degree, spatial relationships, color, multi-object interaction (provider-reported)

The newer Wan 2.2 (July 2025) and Wan 3.0 (Aug 2026) succeed it; Wan 3.0 moved to closed weights, making 2.1 the last fully open generation.

VBench overall 86.22% — provider-reported at launch (Feb 2025); claimed #1 including over Sora, Minimax, and Luma. No independent VBench re-run verified.