Grok Voice Transcribe 2.0 Is Here: Twice the Accuracy, $0.10/Hour, and Automatic Speaker Identification
xAI officially announced 'Grok Voice Transcribe 2.0' on September 18, 2026. Accuracy is roughly double that of version 1.0, available at about $0.10 per hour (batch processing). Speaker diarization — the ability to automatically distinguish between multiple speakers — is included at no extra charge.
Wouldn't it be great if your meeting recordings could automatically turn into text, eliminating the hassle of writing up minutes by hand?
On September 18, 2026, xAI — the company behind Grok — officially announced "Grok Voice Transcribe 2.0." Accuracy has roughly doubled compared to the previous version, the price stays the same, and a new feature automatically distinguishes between multiple speakers in a recording.
What you'll learn in this article
- What changed in Grok Voice Transcribe 2.0
- How much it costs (with approximate USD figures)
- What situations it's most useful for
What Is This?
In a nutshell, it's an AI service that automatically converts recorded audio into text.
For example, you record a one-hour meeting, hand it over to the service, and get back a written transcript of everything that was said. You no longer have to transcribe it manually, saving a huge amount of time on meeting notes.
That said, this service is not something you use directly from a consumer app. It's primarily offered as an API (Application Programming Interface) — a way for developers to connect and integrate it into their own systems. Over time, everyday users will benefit through apps and services built on top of it.
What Changed?
Grok Voice Transcribe 1.0 had a high error rate for short voice commands. There was no way to tell who was speaking, so conversations with multiple people were all output as a single undifferentiated block of text.
The error rate for short voice commands dropped dramatically from 20.6% to 6.8%. The output now automatically labels each speaker — "Speaker A: ..." / "Speaker B: ..." — at no additional cost.
Key Improvements in This Release
Accuracy roughly doubled The error rate for short audio clips (commands like "Yes" or "Next") fell from 20.6% to 6.8% across 20 languages. Overall transcription accuracy for everyday conversations and meetings has also improved.
Speaker diarization added — for free Even in recordings with multiple speakers, the service now automatically identifies who is talking and labels each segment. Output looks like "Speaker A: About yesterday's matter…" with each person's words clearly separated. The fact that this is included at no extra charge is a major plus.
Automatic multi-language detection If the language shifts mid-recording, the system detects and handles it automatically. The service supports dozens of languages, though Japanese support is not officially confirmed in the announcement.
xAI's official announcement does not mention Japanese language support. We recommend testing it yourself to verify accuracy for Japanese audio.
Up to 8 simultaneous channels Multiple microphones or audio sources can be processed at the same time — a feature aimed at recording studios and large multi-participant meetings.
Accuracy, speaker diarization, and multi-language support have all improved, while pricing remains the same as version 1.0. Some estimates put the cost at roughly one-fifth that of a comparable service from OpenAI.
How Can You Actually Use It?
Scenario 1: Automatically create meeting minutes Record a one-hour meeting and send the audio to the API. You get back a transcript broken down by speaker: "Speaker A said X," "Speaker B proposed Y." The work of drafting meeting minutes from scratch is dramatically reduced.
Scenario 2: Transcribe interviews or press recordings Journalists and writers can hand over interview audio and get back text that serves as a ready-made draft. Because speakers are automatically identified, you don't have to sort out "who said what" afterward.
Scenario 3: Add captions to video or audio content You can process YouTube videos or podcast episodes to generate caption data. At roughly $0.10 per hour, even large volumes of content are cost-effective.
How Much Does It Cost?
| Processing Method | Price (per hour) |
|---|---|
| Batch processing (pre-recorded files) | $0.10 |
| Real-time processing (live streams, etc.) | $0.20 |
| Speaker diarization | Free (no extra charge) |
Batch processing means sending pre-recorded files to be transcribed in bulk rather than in real time. For transcription work that isn't time-sensitive, batch processing is the more cost-effective option.
If you're currently using Grok Voice Transcribe 1.0, you can take advantage of 2.0's improved accuracy without changing your code. However, version 1.0 is scheduled to be discontinued in a few weeks, so it's a good idea to verify your migration plan sooner rather than later.
Summary
Grok Voice Transcribe 2.0 is an upgrade that surpasses its predecessor across accuracy, features, and cost. In particular, "free speaker diarization" and "roughly $0.10 per hour" make it highly competitive against other services.
It's especially recommended for use cases like "efficiently producing meeting minutes" or "automating captions for audio content." Once official Japanese-language support is confirmed, this tool will become even more accessible for everyday users.
For full details, see xAI's official announcement page.
Related Articles
More Grok articles
Grok vs ChatGPT: Which Is Better for Work? A Use-Case-by-Use-Case Comparison [Fall 2026]
A head-to-head comparison of Grok 4.7 and ChatGPT (GPT-6 Astra) for everyday work tasks. Grok leads on real-time information and coding cost-efficiency; ChatGPT shines for writing, image generation, and enterprise tool integrations. We break down monthly pricing, each AI's strengths, and tips for choosing the right tool — in plain language.
Grok 4.7 Complete Guide: X Integration, Real-Time Information, and How It Differs from Other AIs
A breakdown of Grok 4.7, previewed by xAI in September 2026. With 2.1 trillion parameters, training on SpaceX data, and real-time integration with X (formerly Twitter), here's what X users need to know — including how it compares to ChatGPT and Claude, explained in plain language.