ByteDance launches Mamoda 2.5, a unified multimodal model.

ByteDance has open-sourced Mamoda2.5, the world's first 25B-level unified multimodal model. Based on the Qwen3-VL-8B architecture, it has 25 billion parameters, with only about 300 million activated at a time. The model achieves state-of-the-art performance across all tasks, including text-to-image generation, text-to-video generation, and video editing. Its inference speed is 12 times faster than Alibaba's Wan2.2 and 18 times faster than Meituan's LongCat Video, with a video editing latency of only 9.2 seconds. The model ranks first in multiple video editing benchmark tests, with performance approaching that of closed-source Sora and Kling.