What was announced
SpaceXSI announced Grok Voice 3.0 on October 14. It is a speech-to-speech model: it receives spoken input and produces spoken replies. The company says it is available through its Voice Agent API, the service developers use to connect voice features to software.
The announcement says applications using grok-voice-latest have already switched to the new model. Official documentation lists voice sessions at $0.08 per minute, checked October 14. This is usage-based access, not unlimited free hosting.
Claims and limits
The launch post cites benchmark results and lower latency. We have not tested the model. Artificial Analysis describes its own tests of reasoning, task completion, and conversation, but a benchmark result does not establish performance on every real call.
Our take for beginners
Editorial analysis: beginners may encounter this model in apps built with the paid service. Voice can help when typing is awkward, but smooth speech still needs fact-checking. Check important details and ask for a written summary. Developers should recheck their applications after an automatic model update.
Source timing
We first covered this October 14 announcement in our afternoon check. The published date does not establish the exact launch hour.