Tencent's Gander aims to keep talking while it works in the background
Tencent's Hunyuan Speech team introduces Gander, a full-duplex voice model pairing a conversational 'cerebellum' with a swappable background agent 'brain'.
Tencent and university researchers unveiled Gander, a model that processes speech, images, and text simultaneously to hold real-time conversations while a background agent handles tasks like coding. A 'cerebellum' manages conversation timing in one-second segments while a swappable 'brain' (tested with an OpenAI GPT-5.6 family model) performs reasoning. Gander achieved the best timing on Full-Duplex-Bench v3, interrupting users in 8% of cases versus 13.5% for GPT-Realtime, but scored slightly below the weakest competitor on task accuracy. It was trained on about 2.7 million examples, and Tencent plans to release the weights and training data after completing its open source process.
- Splits work between a real-time 'cerebellum' and a swappable 'brain' agent like Codex or Claude Code.
- Best conversational timing on Full-Duplex-Bench v3 across all 100 scenarios.
- Task accuracy trails competitors due to speech recognition and perception errors.
- Trained on ~2.7 million examples; weights and training data to be released openly.
- Follows Tencent's July Hy3 open model release and stake talks with agent startup Manus.
- Vendors
- Tencent
From the original article
4 min read at the-decoder.com
Tencent's Gander processes speech, images, and text while handling tasks in the background. A "cerebellum" keeps the conversation going, while a swappable "brain" searches files, writes code, or tackles other complex work. Users can interrupt or change the task mid-conversation.
By Jonathan Kemper · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.