Grok Voice 3.0

// Overview

Grok Voice 3.0 is the speech-to-speech model behind the Voice Agent API from SpaceXAI (formerly xAI), announced on 14 October 2026 and served under the API model id `grok-voice-3.0`. The API itself did not change: the alias `grok-voice-latest` now resolves to 3.0, so voice agents that use the alias were moved to the new model automatically, and `grok-voice-2.0` is marked deprecated in xAI's release notes.

The release is a quality upgrade rather than a new interface. On Artificial Analysis' independent measurements dated 13 October 2026, Grok Voice 3.0 scores 86.9 on the Speech-to-Speech Index, ranked first, against 83.2 for OpenAI's GPT-Live-1 (Astra, Medium) and 82.6 for Gemini 3.8 Live. It also leads on turn-taking and interruptions (97.0% on Full Duplex Bench), speech reasoning (98.5% on Big Bench Audio) and human preference (1070 Elo in the Speech Agent Arena).

The model is aimed at voice agents that resolve real calls. SpaceXAI says it runs in Tesla vehicles for hands-free requests and powers Grok Bot's voice mode, and that Starlink's Grok Voice agent answers about 20,000 support calls a day in several languages, resolving roughly 70% without a human. On ServiceNow's open EVA-Bench, which covers more than 200 enterprise voice scenarios, it has the highest accuracy at 74%.

Pricing is unchanged from Grok Voice 2.0: $0.08 per minute of audio, or $4.80 per hour, in the Voice Agent API. On Big Bench Audio, Artificial Analysis puts its cost per task at $0.025, against $0.030 for GPT-Live-1 and $0.055 for GPT-Realtime-2.1 (High).

// Benchmarks

This model's scores

1. AA Speech-to-Speech Index86.9%
2. Speech Reasoning (Big Bench Audio)98.5%
3. Conversational Dynamics (Full Duplex Bench)97%
4. Task Success Rate94.8%
5. Agentic Performance (τ-Voice)71.6%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Billed by audio time: $0.08 per minute ($4.80 per hour) in the Voice Agent API, unchanged from Grok Voice 2.0.

Pricing source ↗

// Strengths

• 86.9 on the Artificial Analysis Speech-to-Speech Index — ranked first, ahead of GPT-Live-1 (83.2) and Gemini 3.8 Live (82.6)
• 97.0% on Full Duplex Bench for handling pauses, turn-taking and interruptions
• Highest accuracy on ServiceNow's EVA-Bench at 74% across 200+ enterprise voice scenarios
• Drop-in upgrade: same API and price as 2.0, and `grok-voice-latest` sessions moved automatically
• Lowest cost per Big Bench Audio task of the models compared, at $0.025

// Best for

• Reach for it for customer-support and sales phone agents that need to resolve a call end to end, not just transcribe it.
• Reach for it when callers interrupt, pause and talk over the agent — its strongest published score is on full-duplex turn-taking.
• Reach for it to upgrade an existing Grok Voice agent without code changes, by pinning `grok-voice-3.0` or staying on the alias.
• Reach for it for in-car or hands-free assistants, the deployment SpaceXAI describes in Tesla vehicles.

// How to access

// Grok Voice (speech-to-speech) — every version

The full lineage of the Grok Voice (speech-to-speech) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.