AI news · October 15, 2026
xAI makes Grok Voice 3 available through its API
A low-latency voice agent is now a metered API product, so builders can test a real call workflow before committing to a full telephony stack.
- speech-to-speech price
- $0.08/min
- latency claim
- <1s
- languages
- 25+
xAI announced Grok Voice 3 on October 14 as a speech-to-speech model available through its API. The service supports full-duplex conversation, tool calls, search, multi-turn context and expressive speech. The company lists sub-second latency, more than 25 languages and a speech-to-speech price of $0.08 per minute. The API page also exposes text-to-speech, speech-to-text, speaker diarization and custom-voice features.
The useful product change is packaging: a builder can measure a voice workflow with the same API surface used for a production agent, rather than stitching together separate speech services. xAI also reports benchmark results, but those are company-published claims. Treat task completion, transfer rate, latency and cost per resolved call as the real acceptance tests.
What you can do with it
Try it: build one narrow support or sales flow in the Grok Voice API. Run 20 short calls with a fixed script, a tool call and a human handoff. Record first-response latency, completion rate, transfer rate and total speech cost before expanding the workflow.
Our take
This is most useful when conversation speed matters more than model novelty. The price supports a bounded canary, but voice quality does not prove business value. A measured call-resolution test should decide whether the API replaces an existing speech stack.
Links Try itRelease notes
Source: xAI — Made With Models writes the brief; the reporting is theirs.