Tencent's Gander handles conversations while tackling tasks

tencent's gander model is designed for real-time conversations while handling complex tasks in the background. it processes speech, images, and text simultaneously, with a 'cerebellum' managing conversations and a swappable 'brain' handling agent tasks. in tests, gander interrupted users less often than competing models but showed weaknesses in task accuracy and video/audio understanding. the team plans to release model weights and training data. gander's design addresses the tradeoff between conversational speed and reasoning ability.