Tencent Unveils WeMM-Video Models for Video Understanding
The Chinese tech giant released four open-source models spanning 2 billion to 35 billion parameters for multimodal video, text, image, and audio tasks.
Updated on Nov. 20, 2026 in Artificial Intelligence

Tencent introduced WeMM-Video, a Chinese-English model family that handles video, text, image, and audio understanding. The family comes in four versions, and the company published the models in a public GitHub repository for developers and researchers.
Why it matters
The release gives developers and researchers openly available tools for multimodal AI, a field where proprietary systems have largely led. Making the models public lets outside teams test, reproduce and build on the results.
The four models span 2 billion to 35 billion parameters and were trained on a corpus of about 1.2 trillion tokens, with each supporting both an instruction mode for direct answers and a thinking mode for step-by-step reasoning. The developer's self-reported evaluations across more than 120 benchmarks are not yet independently confirmed.
The players
Tencent
Tencent is a major Chinese technology conglomerate known for its social media, gaming and cloud businesses, and it has invested heavily in artificial intelligence research.
The details
Tencent published the WeMM-Video family in a GitHub repository that includes an evaluation and reproduction pipeline along with guides and examples in English and Simplified Chinese. The models are designed for omni-modal understanding, processing video, text, image and audio inputs together.
Timeline
November 20, 2026: Tencent introduces the WeMM-Video model family and publishes it on GitHub.
The Tech Race
WeMM-Video enters a competitive open-source field where Alibaba's Qwen3-Omni series set a benchmark for omni-modal models, and Tencent's developer claims a smaller model beats the larger Qwen3-Omni variant. The release adds Chinese labs' open multimodal offerings to a race increasingly contested against proprietary systems such as Gemini-3.0-Flash.
Developers and researchers can download the WeMM-Video models and evaluation pipeline from GitHub to run them locally or in their own projects. The models are released openly, so using them does not require a commercial subscription to a proprietary platform.
The takeaway
Open-source multimodal models are narrowing the gap with proprietary systems, so teams building video or audio AI tools can now evaluate capable alternatives directly. Checking reported benchmark claims against your own tasks remains essential before adopting any new model.
Further reading
For more on the broader field, explore our Artificial Intelligence coverage.
More information
Access the models, evaluation pipeline and guides in Tencent's WeMM-Video GitHub repository.
Source note: This article includes information reported by TokenPost.


