WeMM-Video-35B-A3B
AvailableFlagship of Tencent's WeMM-Video family of Chinese-English bilingual omni-modal models for joint audio-visual understanding and reasoning: a mixture-of-experts model with 35B total and 3B active parameters (Qwen3.5-MoE-architecture backbone, 40 layers, 256 experts with 8 active per token per config). It takes text, image, audio and video input (including video with its original soundtrack) and outputs text, with instruct and thinking modes. Pretrained on about 1.2T multimodal tokens, then post-trained with multi-teacher OPD, anti-repetition MrDPO and specialist RL. Tencent claims a new open-source state of the art in omni-modal video understanding, ahead of Qwen3.5-Omni-Flash and Seed2.0-Mini-0428 and closing in on Gemini-3.0-Flash (vendor-reported, over 120 benchmarks). Technical report on arXiv Nov 19 2026; Apache 2.0 weights uploaded to Hugging Face Nov 20 2026 (about 36.2B parameters in the checkpoint, including encoders), with FP8, AWQ, GGUF and NVFP4 builds. Context length not stated on the model card.
Specifications
- License
- Open source · Apache-2.0
- Weights
- Downloadable
- Architecture
- Mixture-of-Experts
- Parameters
- 35B · 3B active
- Context window
- — tokens
- Max output
- —
- Knowledge cutoff
- —
- Price (in / out, $/M)
- —
- Modalities
- TextVisionAudioVideo
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.