Tencent Ships Gander, an Open Omni-Interaction Agent With 100% Turn-Taking Accuracy on Full-Duplex Bench
Summary
Tencent's Hunyuan Speech team released Gander, a 9B end-to-end omni-interaction agent that fuses streaming audio, video, and text into one 1-second-chunk token flow. A 'front cerebellum' handles real-time turn-taking (100% appropriate vs 96% for GPT-Realtime, 8% premature starts vs 13.5%) while a plug-in 'back brain' runs longer reasoning via Claude Code or Codex. Weights, code and training data will be released.
Originally reported by huggingface.co
Read the original article →Original headline: Tencent Ships Gander, an Open Omni-Interaction Agent With 100% Turn-Taking Accuracy on Full-Duplex Bench