AI CURATED HOT STORY · 11/11/2026
Fish Audio has officially launched Live 3, a full-duplex voice model that internalizes turn-taking capabilities into the speech model itself. Instead of relying on silence thresholds and hand-written rules, it learns from real human conversations when to acknowledge, wait, respond, or stop. The company claims a response time as fast as 186 milliseconds in short interjection scenarios, but the initial release only supports English and no benchmark scores have been published.
The model supports simultaneous listening and speaking, naturally inserting feedback like “mm-hmm” and stopping at appropriate moments. Users can generate a working voice agent with just one sentence. Live 3 is now available on the Fish Agent Platform, API, and enterprise on-premises deployment, priced at $0.035 per minute until the end of 2026. Fish Audio also offers custom enterprise agent bookings and free builds for the first 1,000 use-case submissions.
Live 3 marks a key step from walkie-talkie-style turn-taking toward natural full-duplex conversation. Internalizing turn-taking into the model could accelerate voice agent adoption in customer service, companionship, and real-time translation. However, English-only support and the lack of benchmark scores indicate the product is still in an early stage and requires further iteration.
The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.