In brief
- SpaceXAI launched Grok Voice 3.0, a speech-to-speech model now live in its Voice Agent API.
- It ranks first on the Artificial Analysis Speech-to-Speech Index, with replies in under one second.
- The price stays at $0.08 per minute, the same as Grok Voice 2.0.
SpaceXAI has launched Grok Voice 3.0, a speech-to-speech model for voice agents that developers can use today for $0.08 per minute.
The company announced it on its official website on October 14, 2026. SpaceXAI says the model replies in under one second and costs less to run than other frontier voice models.
What is new in Grok Voice 3.0
SpaceXAI calls Grok Voice 3.0 the most intelligent voice model in the world. The figures below come from Artificial Analysis testing, as cited in the announcement:
| Benchmark | Grok Voice 3.0 | Grok Voice 2.0 | GPT-Live-1 | Gemini 3.8 Live |
|---|---|---|---|---|
| Speech-to-Speech Index (overall) | 86.9 | 81.3 | 83.2 | 82.6 |
| Support-call success (τ-Voice) | 71.6% | 56.5% | 74.5% | 68.6% |
| Natural turn-taking (Full Duplex Bench) | 97.0% | 95.1% | 94.9% | 91.9% |
On τ-Voice, which tests support calls with accents, background noise and interruptions, OpenAI’s GPT-Live-1 still scores higher. SpaceXAI says Grok Voice 3.0 comes close with 40% lower latency.
Cost is another gap. Per Artificial Analysis, an average task costs $0.025 on Grok Voice 3.0, compared with $0.030 on GPT-Live-1 and $0.055 on GPT-Realtime-2.1.
On ServiceNow’s EVA-Bench benchmark for enterprise voice agents, the model posted the top accuracy score, 74%.
Where it already runs
SpaceXAI says the model handles millions of conversations a day in Tesla vehicles. Drivers ask it to check a calendar or answer an email, and through Grok Bot to order a coffee or book dinner.
Grok Voice 3.0 also powers Grok Bot’s voice mode. In Starlink call centers it answers 20,000 calls a day in multiple languages, and about 70% are resolved with no human involved.
Why it matters
For a business, latency and price per minute decide whether a voice agent can take over part of a call center. The price did not rise from the previous generation, so switching to the new model does not change the bill.
| Model | Per minute | Per hour | Status |
|---|---|---|---|
| Grok Voice 3.0 | $0.08 | $4.80 | Flagship |
| Grok Voice 2.0 | $0.08 | $4.80 | Deprecated |
Developers using grok-voice-latest have already been switched over automatically, with no code changes. Anyone who pinned grok-voice-2.0 needs to move to grok-voice-3.0 manually.
SpaceXAI picked the benchmarks for its own announcement, so real-world call quality is the test that matters.
What we don’t know yet
- Which languages the benchmark results were measured in. The docs list 20+ supported languages.
- Whether every Grok Bot user already gets the new model in voice mode, or if it is rolling out gradually.
- Whether the consumer Grok app also moves to Grok Voice 3.0. The announcement covers the API, Tesla and Grok Bot.
What it means for you
- Developers: Grok Voice 3.0 is in the Voice Agent API at $0.08 per minute. If you pinned grok-voice-2.0, update the model name. New to the term? See API in our glossary.
- Businesses: before replacing human agents, test the voice agent on real calls with names, phone numbers and addresses.
- Grok Bot users: the bot’s voice mode runs on the new model. Our Grok Bot language support guide explains voice chat and its language setting.
- New to Grok Bot? Our complete Grok Bot guide covers plans, pricing and setup.
More on Maitz
Sources
- SpaceXAI: Introducing Grok Voice 3.0, official x.ai website, October 14, 2026
- SpaceXAI: official Voice Agent API documentation (models and supported languages), docs.x.ai
- Artificial Analysis, as cited in the official announcement, October 13, 2026
- Verified: October 14, 2026
