Last released Jul 17, 2026
Use interim transcripts to cut voice-to-voice latency — KV-cache prefilling, speculative generation, intent extraction, and tool prefetch while the user is still talking.
Supported by
