Wan 2.6

• - 🏢 Alibaba's Wan video family; text-to-video generation, released December 2025.
• - 👁️ Photorealistic output with strong subject coherence across frames.
• - 💬 Produces video from natural-language prompts as temporally linked frames.
• - 🆕 Part of a broader Wan 2.6 lineup spanning multiple video tasks.
• - 🌐 Accessible via Alibaba Cloud Model Studio's DashScope SDK.
• - 🆕 Successor Wan 2.7 extends to first-and-last-frame and video continuation.
• - 📚 No independently verified benchmarks from primary or top evaluators available.

Alibaba

Alibaba Group is a Chinese multinational technology company founded in 1999 and headquartered in Hangzhou, Zhejiang. Originally built around e-commerce and cloud computing, Alibaba has become one of the most prolific contributors to open-weight AI research,…

Wan 2.6 is part of Alibaba's Wan family of video generation models, developed under the company's AI research division. This text-to-video entry, released in December 2025, takes a natural-language prompt and produces photorealistic clips with strong subject coherence across frames, rendering video as a sequence of temporally linked frames. It sits within a broader Wan 2.6 lineup that also covers image-to-video and related video tasks.

The model follows Alibaba's earlier Wan releases, including the Wan 2.5 Preview and the open-weight Wan 2.2 A14B generation, both of which appear as siblings in the catalog. These represent the prior steps in the family's text-to-video progression leading up to Wan 2.6.

The line has since advanced to Wan 2.7. Alibaba Cloud's documentation notes that Wan 2.6 and earlier image-to-video variants only support first-frame-to-video conditioning, whereas Wan 2.7 extended that capability to first-and-last-frame control and video continuation. This marks the most concrete documented generational difference between Wan 2.6 and its direct successor.

Wan 2.6 remains accessible through Alibaba Cloud Model Studio's DashScope SDK, making it a practical option for creators generating short clips from text prompts. No independently verified benchmark scores for this specific model were available from primary or top-evaluator sources at the time of writing.