# Grok Voice Think Fast 2.0

Grok Voice Think Fast 2.0 powers voice agents that listen and reply out loud. Audio streams both ways over a WebSocket session, and Grok Voice Think Fast 2.0 reasons while it speaks so tool calls fire early in the reply.

- **Model ID:** `spacexai/grok-voice-think-fast-2.0`
- **Type:** realtime
- **Providers:** xai
- **Context window:** Not applicable
- **Maximum output tokens:** 0
- **Pricing:** See the provider details below
- **Canonical page:** https://vercel.com/ai-gateway/models/grok-voice-think-fast-2.0

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { gateway } from '@ai-sdk/gateway';

export async function POST() {
  const { token, url } = await gateway.experimental_realtime.getToken({
    model: 'spacexai/grok-voice-think-fast-2.0',
  });

  return Response.json({ token, url, tools: [] });
}
```

## About

Grok Voice Think Fast 2.0 is SpaceXAI's realtime voice model, released July 29, 2026 and available through AI Gateway. Audio goes in and audio comes back, so Grok Voice Think Fast 2.0 handles full-duplex conversation without a chain of separate transcription, text, and speech models. SpaceXAI reports improvements over the previous Grok Voice model in reasoning, transcription accuracy, and conversation.

Grok Voice Think Fast 2.0 reasons in parallel with speech, working through a query while the conversation keeps moving. Grok Voice Think Fast 2.0 also spends fewer reasoning tokens than the previous Grok Voice model, so tool calls fire sooner, often before the agent finishes its first sentence. Transcription holds up in real-world conditions, including background noise and telephony compression.

Through AI Gateway, a server route mints a short-lived token so your API key never reaches the client, and the `useRealtime` hook from `@ai-sdk/react` drives the WebSocket, microphone capture, and playback from the browser, with session config for voice, turn detection, instructions, and tools. Realtime support is in beta and ships on the canary releases of the AI SDK.

## What to consider

Grok Voice Think Fast 2.0 supports speech-to-speech only. Realtime sessions don't transcribe or translate recorded files, so use `grok-stt` for that. Image input isn't supported in realtime sessions either.

AI Gateway enforces session limits: 25 minutes maximum duration, a five-minute idle timeout, a 30-second window for the first client message, and a 256 KB cap on message size. Teams also have a limit on concurrent sessions. Reconnecting doesn't resume a previous session, so replay any context you need after a drop. Realtime support is in beta and ships on canary AI SDK releases, so expect API changes.

## When to use

### Best For

- **Realtime voice agents** holding full-duplex conversation with spoken replies while the user talks
- **Phone and support lines** needing transcription that holds up under background noise and telephony compression
- **Early tool calling** where lower reasoning overhead fires tool calls within the first spoken sentence
- **Reasoning during conversation** so parallel thinking works through hard queries while speech continues
- **Grok Voice upgrades** stepping up from Grok Voice Think Fast 1.0 on the same realtime surface

### Consider Alternatives When

- **Recorded audio transcription** is the task, where `grok-stt` transcribes files and realtime sessions are speech-to-speech only
- **One-way speech output** is enough, where `grok-tts` generates narration without a live session
- **Long uninterrupted sessions** are required, since AI Gateway caps realtime sessions at 25 minutes
- **Text-only conversations** carry the workload, where a text model such as Grok 4.5 costs less with no audio involved

## Best for

- **Realtime Voice Agents:** Full-duplex conversation with spoken replies while the user talks
- **Phone And Support Lines:** Transcription that holds up under background noise and telephony compression
- **Early Tool Calling:** Lower reasoning overhead fires tool calls within the first spoken sentence
- **Reasoning During Conversation:** Parallel thinking works through hard queries while speech continues
- **Grok Voice Upgrades:** A step up from Grok Voice Think Fast 1.0 on the same realtime surface

## Consider alternatives

- **Recorded Audio Transcription:** `grok-stt` transcribes files, since realtime sessions are speech-to-speech only
- **One-Way Speech Output:** `grok-tts` generates narration without a live session
- **Long Uninterrupted Sessions:** AI Gateway caps realtime sessions at 25 minutes, so plan reconnect logic
- **Text-Only Conversations:** A text model such as Grok 4.5 costs less when no audio is involved

## Frequently asked questions

### How do I connect to Grok Voice Think Fast 2.0 through AI Gateway?

Add a server route that mints a short-lived token with `gateway.experimental_realtime.getToken`. In the browser, the `useRealtime` hook from `@ai-sdk/react` handles the WebSocket connection, microphone capture, and playback. In Node.js, drive a WebSocket yourself using the realtime model as a codec.

### How does Grok Voice Think Fast 2.0 improve on Grok Voice Think Fast 1.0?

SpaceXAI reports gains in reasoning, transcription accuracy, and conversation. Grok Voice Think Fast 2.0 also spends fewer reasoning tokens, so tool calls fire sooner in a reply.

### What does it mean that Grok Voice Think Fast 2.0 reasons in parallel with speech?

Grok Voice Think Fast 2.0 works through a query while it keeps talking, rather than pausing to think and then answering. Replies stay responsive on complex, multi-step requests.

### Can Grok Voice Think Fast 2.0 transcribe or translate recorded audio?

No. Grok Voice Think Fast 2.0 supports speech-to-speech only. Use `grok-stt` to transcribe recorded files and `grok-tts` to generate narration from text.

### Does Grok Voice Think Fast 2.0 accept image input?

No. Realtime sessions on AI Gateway don't support image input. Send visual context to a separate vision model and pass the result into the session as text.

### What session limits apply to Grok Voice Think Fast 2.0?

AI Gateway caps sessions at 25 minutes, closes idle sessions after five minutes, expects a first client message within 30 seconds, and rejects messages over 256 KB. Teams also have a concurrent session limit.

### How do I authenticate with Grok Voice Think Fast 2.0 through Vercel AI Gateway?

Use your Vercel AI Gateway API key on the server with `spacexai/grok-voice-think-fast-2.0` as the model identifier. `getToken` exchanges the key for a single-use, short-lived client secret, so the key never reaches the browser.

### What does Grok Voice Think Fast 2.0 cost?

See the pricing section on this page for today's rates. AI Gateway reflects each provider's pricing for Grok Voice Think Fast 2.0 with no markup.

### Does Vercel AI Gateway support Zero Data Retention for Grok Voice Think Fast 2.0?

Yes, Zero Data Retention is available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

## Links

- [Model page](https://vercel.com/ai-gateway/models/grok-voice-think-fast-2.0)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://x.ai/voice)


If you encounter incorrect information, can’t find what you need, need a capability Vercel doesn’t currently offer, or have any other feedback to report, please do it by following the instructions at https://vercel.com/feedback.md.