Native interaction models cut latency to 0.4s
Thinking Machines achieves 0.4s turn-taking latency by training from scratch for concurrent listening and speaking, removing legacy speech wrappers.
Priya Nair
11 min read