Tencent Releases WeMM-Video: Omni MLLMs from 2B to 35B-A3B
ExponentialCookie · reddit · 2026-11-20
Tencent released WeMM-Video, a family of omni multimodal LLMs in accessible sizes (2B, 4B, 9B, 35B-A3B), with the paper on arXiv (2611.26409).
The models combine video frames with the original soundtrack to describe events, transcribe speech, and reason over what is seen and heard; they also accept silent video, standalone audio, images, and text, responding in Chinese or English. Vision and audio encoders turn media into tokens processed by the LM, with timestamped video frames. Pretraining used 1.2 trillion multimodal tokens, followed by multi-teacher OPD, anti-repetition MrDPO, and specialist RL.
Related event: Tencent and Tsinghua Open-Source WeMM-Video Omni-Modal Video Models(2 posts)→
More from Multimodal
- "AGI Before GTA6": AI-Generated Music Video Riffs on 13-Year GTA VI Wait — Yamapama · 2026-11-21
- Meta's AVTR adds visual imagination to multimodal reasoning, +4.5 pts across 16 benchmarks — liuzhuang1234 · 2026-11-21
- Genie 4 world model live on Comfy playground, early testers impressed — EnvironmentalDay7419 · 2026-11-21
- Roblox-style AI horror short built with CUT + local ComfyUI, workflow shared — rileygstaliger · 2026-11-21
- AI-made 'The Office' crossover puts Michael Scott across from Mr. Bean in a hilariously faithful clip — alanskimp · 2026-11-21
- AI Song Contest 2026 finalists announced; two Stable Audio-affiliated teams made the top 14 — jordiponsdotme · 2026-11-21