Overview
Grok Voice 3.0 is the speech-to-speech model xAI released on October 14, 2026. Audio goes in and audio comes out, without a transcription step in the middle. xAI positions it as the voice layer for real conversations rather than a text-to-speech reader, with customer support and voice agents as the target workloads.
In independent testing published on October 13, 2026, Artificial Analysis ranked it first on the Speech-to-Speech Index with a score of 86.9. The index combines speech reasoning, tau-Voice, and preference and task success from blind head-to-head calls. Broken out, Big Bench Audio sits at 98.5%, Speech Agent Arena Elo at 1070, task success rate at 94.8% and Full Duplex Bench at 97%. It scores 71.6% on tau-Voice, which xAI describes as comparable with GPT-Live-1 at its Astra setting while running at 40% lower latency.
Two things give it scale on day one: Grok Voice already carries millions of conversations a day across Tesla and SpaceX, and the model landed in the API and in Voice Agent Builder on release day at the same price as its predecessor.
Key Features
- Direct speech to speech: Audio in, audio out, with no transcription in the middle
- First on the Artificial Analysis index: 86.9 on the Speech-to-Speech Index, tested by Artificial Analysis on October 13, 2026
- Full duplex conversation: 97% on Full Duplex Bench, handling pauses, turn-taking and interruptions at a natural pace
- Under one second to first audio: The latency figure xAI quotes, with roughly 40% lower latency than comparable settings in tau-Voice testing
- 94.8% task success: Task completion in Artificial Analysis blind head-to-head calls
- API and Voice Agent Builder: Available on release day through xAI APIs and the visual builder
- Running at Tesla and SpaceX: xAI reports millions of conversations a day across both product lines
- Predictable cost per task: Lower average cost per task on the Artificial Analysis Big Bench Audio run than GPT-Live-1 and GPT-Realtime-2.1
Use Cases
- Contact centres that let a voice agent take the first pass on inbound calls
- Voice assistants that need real-time interruption and turn-taking
- Extending an existing text support script to the phone channel
- Low-latency voice interaction inside cars and hardware
- Automated conversations for outbound calling and follow-ups
Pros
- Direct speech-to-speech output removes a transcription stage and its error surface
- Ranked first on the Speech-to-Speech Index in independent Artificial Analysis testing
- Shipped into the API and Voice Agent Builder on release day, so the integration path is clear
- Priced the same as the previous generation, so upgrading needs no new cost model
- Already carrying millions of conversations a day across Tesla and SpaceX
Pricing
The model is served through the xAI API and billed on usage. xAI says pricing matches the previous generation, and community figures put it around USD 0.08 per minute. Exact rates and quotas follow the official xAI pricing page.
Summary
Grok Voice 3.0 is the speech-to-speech model xAI released on October 14, 2026, taking audio in and returning audio directly. Independent testing by Artificial Analysis on October 13, 2026 put it first on the Speech-to-Speech Index at 86.9, with Big Bench Audio at 98.5%, Speech Agent Arena Elo at 1070, task success at 94.8% and Full Duplex Bench at 97%; time to first audio is under a second, and it scores 71.6% on tau-Voice at roughly 40% lower latency than comparable settings. It already carries millions of conversations a day across Tesla and SpaceX, shipped into the API and Voice Agent Builder on release day, and costs the same as the generation before it. It fits phone support, voice assistants and outbound calling; for plain narration or voiceover, a real-time conversation model is more than the job needs.
Version History
- Grok Voice 3.0 launches, first on the Artificial Analysis Speech-to-Speech Index at 86.9 (2026-10-14): On October 14, 2026, xAI released Grok Voice 3.0, which it calls the most intelligent speech-to-speech model. In independent testing by Artificial Analysis dated October 13, 2026, Grok Voice 3.0 ranks first on the Speech-to-Speech Index at 86.9, ahead of Grok Voice 2.0 at 81.3, GPT-Live-1 at its Astra setting at 83.2 and Gemini 3.8 Live at 82.6. Component scores include Big Bench Audio at 98.5%, Speech Agent Arena Elo at 1070, task success rate at 94.8% and Full Duplex Bench at 97%. It scores 71.6% on tau-Voice, which xAI describes as comparable with GPT-Live-1 at its Astra setting at 40% lower latency, with time to first audio under one second, and a lower average cost per task on the Big Bench Audio run than GPT-Live-1 and GPT-Realtime-2.1. The model reached the API and Voice Agent Builder on release day at the same price as its predecessor, and already carries millions of conversations a day across Tesla and SpaceX.