Grok Bot starts talking back on desktop and mobile

The post announcing voice mode for Grok Bot signs off with five words: "As it has been foretold." The mode reaches desktop and mobile apps gradually over the next couple of days, according to the SPACEXAI post on Threads.

At a glance

• Voice calls for Grok Bot were already in testing as of August 26, and the announcement now moves the mode out of testing into a staged rollout on desktop and mobile.
• Basenor reports voice mode in both the desktop and the mobile apps, and SPACEXAI attaches only one time frame to it: gradually, over the next couple of days.
• Not stated: which subscription tiers get voice first, what gradually means in practice, or how the feature works under the hood on either platform.

If you missed the earlier episodes: according to ScriptByAI, Grok Bot entered early beta on August 11, 2026 as an autonomous AI teammate that can sign in to tools and finish work in the background. The same account says access widened to all SuperGrok Plus, Cursor Pro+ and Cursor Teams subscribers on August 22, 2026, and to all SuperGrok and Cursor Pro subscribers on August 26, 2026.

SPACEXAI gives the rollout a couple of days

SPACEXAI put the schedule plainly:

The feature will be rolling out gradually over the next couple of days.

Voice calls for Grok Bot were in testing as of August 26. Basenor reports the mode in both the desktop and the mobile apps, so it is not phone-only. The SPACEXAI account calls voice an enabler for loads of new use cases and asks whether people have it yet.

A staged rollout normally means the app build is the same for everyone and the switch is flipped account by account. If there is no call button in your app today, that is the ordinary state on day one of a gradual release.

Where did the voice plumbing come from?

According to ScriptByAI, SpaceX announced its acquisition of xAI on February 2, 2026, SpaceXAI announced standalone speech-to-text and text-to-speech APIs on April 17, 2026, and on April 30, 2026 Custom Voices added voice cloning for the Grok TTS and Voice Agent APIs. By August 2026 the same account describes a product family spanning Grok 4.6, Grok Bot, Grok Build, Grok Imagine, voice APIs, Office add-ins and enterprise integrations.

ScriptByAI also dates Grok 4.6 to August 12, 2026, launched for long-running agents, complex coding and knowledge work across Grok Build, Cursor, Grok Bot and the SpaceXAI API, and names it as the model under Grok Bot at the time voice mode began rolling out.

Swipefile published a 2025 rollout roadmap graphic for Grok and xAI products that places "new voice modes" inside a Q1 feature cluster. Grok's own release notes, separately, describe Grok 4.3 as a pre-trained model with an improved architecture and a December 2025 knowledge cutoff, rolling out to SuperGrok and Premium+ subscribers.

What has to happen between your question and Grok's answer?

Real-time voice assistants generally run as a streaming pipeline. Retellai describes three stages: speech-to-text turns your audio into text, a large language model reads that text plus context and tools and decides on a reply, and text-to-speech turns the reply back into audio. By Retellai's account the full loop has to close in under roughly 700 milliseconds to feel natural.

Retellai says the speed comes from streaming at every stage instead of waiting: speech-to-text emits partial transcripts roughly every 50 milliseconds, the model streams tokens as they are generated, and text-to-speech starts streaming audio chunks before the full reply text exists. A simultaneous interpreter works the same way, talking before your sentence ends.

Two more components wrap that core, again by Retellai's account: turn-taking detection, which decides when the caller has stopped speaking, and barge-in handling, which decides what to do when the caller cuts in. That, it says, is where most of the latency and the engineering effort actually sits.

The post does not name which tiers see voice first and says nothing about how the mode is built, so the pipeline above is how such systems work in general, not a description of Grok's. ScriptByAI describes Grok Bot as a teammate that works in the background; in our view a live call is the more interesting shift.

Who gets voice next

One time frame is on the record, a couple of days from the post, and nothing beyond it. The tiers are not named, and nothing states that voice follows the order in which Grok Bot's own access widened in late August. There is no stated end date for the rollout and no list of supported platforms beyond desktop and mobile. Until that changes, the only check is opening the app and looking for the call button.