Grok Voice Think Fast 1.0

// Overview

Grok Voice Think Fast 1.0 is the first model in xAI's Think Fast voice line, announced on 23 April 2026 and served through the Speech to Speech API as `grok-voice-think-fast-1.0`. xAI positions it for complex, ambiguous, multi-step workflows across customer support, sales and enterprise operations, and specifically for high-stakes calls that demand precise data entry and high-volume tool calling.

Its defining trick is that reasoning happens in the background: the model can think through a difficult query or workflow mid-call without that deliberation surfacing as extra response latency. xAI says it natively supports 25+ languages, which it frames as the reason the model suits global deployments rather than one market at a time.

At launch xAI reported that it topped the τ-voice Bench leaderboard — an evaluation of full-duplex voice agents under realistic conditions including background noise, accents, interruptions and turn-taking, run across retail, airline and telecom scenarios — ahead of Grok Voice Fast 1.0, Gemini 3.1 Flash Live and GPT Realtime 1.5. xAI also published production numbers from its Starlink deployment: a 20% conversion rate on sales inquiries and a 70% resolution rate on customer support, with 28 distinct tools wired across hundreds of workflows.

It was superseded as the line's default by Grok Voice Think Fast 2.0: xAI's release notes record the `grok-voice-latest` alias routing to 2.0 from 5 August 2026. Version 1.0 is not marked deprecated in those notes and is still addressable by its explicit model id. In the comparison xAI published with 2.0, version 1.0 scores 75.7% on the AA Speech-to-Speech Quality Index, 97.1% on Big Bench Audio, 77.8% on Full Duplex Bench and 52.1% on τ-voice Bench, at 1.25 seconds to first audio.

// Benchmarks

1. AA Speech-to-Speech Quality Index75.7%
2. Speech Reasoning (Big Bench Audio)97.1%
3. Conversational Dynamics (Full Duplex Bench)77.8%
4. Agentic Performance (τ-voice Bench)52.1%

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

// Strengths

• Background reasoning during a call, so deliberation on a hard query does not add response latency
• 25+ natively supported languages, aimed at multi-market deployments
• Topped the τ-voice Bench leaderboard at launch, which tests full-duplex agents under noise, accents and interruptions
• Built for high-volume tool calling — xAI's own Starlink deployment wired in 28 distinct tools across hundreds of workflows

// Best for

• Reach for it when an existing integration is pinned to the 1.0 model id and you need its documented behaviour rather than 2.0's.
• Reach for it for support and sales lines where the agent must capture data accurately and call many tools per conversation.
• Reach for it as the baseline when benchmarking what Think Fast 2.0 changed — xAI publishes both columns side by side.

// 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.