xAI released Grok Voice Transcribe 2.0, a new speech-to-text model the company says is twice as accurate as its predecessor while keeping the exact same per-minute price. The update lands as more creators, podcasters, and businesses lean on automatic transcription for captions, show notes, and voice agents, and it is the kind of quiet infrastructure update that shows up in our daily AI news roundups more than in headlines.
xAI says the new model ranks first for accuracy among 32 streaming speech-to-text systems tracked on the Artificial Analysis streaming leaderboard, with the biggest gains showing up in noisy, real-world audio and multilingual conversations. The model is already rolling out through xAI’s Speech-to-Text API, and it also powers voice features inside Grok and in Tesla vehicles. Here is what actually changed, what it costs, and who can use it today.
What’s New?
According to xAI’s official announcement, Grok Voice Transcribe 2.0 was trained to handle the kind of audio that trips up older transcription models: phone-line calls, overlapping speakers, heavy accents, and spoken credentials like account numbers or confirmation codes. xAI reports the model is “twice as accurate” as Voice Transcribe 1.0 at the same price, and that on short multilingual phrases, word error rate dropped from 20.6 percent to 6.8 percent.
The model also gained automatic language detection with the ability to switch languages mid-recording, which matters for creators who mix languages in a single video or livestream, a common pattern for AI tutorial content aimed at global audiences.
Previous Version vs New Version
| Detail | Voice Transcribe 1.0 | Voice Transcribe 2.0 |
|---|---|---|
| Multilingual word error rate (short phrases) | 20.6% | 6.8% |
| Speaker diarization | Limited | Included at no extra cost |
| Multichannel audio | Not supported | Up to 8 channels |
| Key term biasing | Not available | Up to 100 terms per request |
| Pricing (batch / streaming) | $0.10 / $0.20 per hour | $0.10 / $0.20 per hour (unchanged) |
| Status | Being phased out over “coming weeks” | Becoming the new default |
Features
Beyond the accuracy claims, xAI’s release notes and independent coverage from Unite.AI point to a fuller feature set built for both creators and developers:
- Batch and real-time streaming transcription through the Speech-to-Text API
- Word-level timestamps with confidence scores
- Speaker diarization included at no extra cost
- Support for up to 8 audio channels in a single file
- Key term biasing, so names, brands, or technical terms get transcribed correctly
- Automatic text formatting for numbers, dates, currencies, and emails
- Filler word removal and smart turn detection, aimed at voice agents
- Support for a dozen audio formats and files up to 500 MB
Availability
Grok Voice Transcribe 2.0 is available now through xAI’s Speech-to-Text API. Existing integrations reportedly get the upgrade automatically without code changes, since it sits behind the same endpoint used by Voice Transcribe 1.0. The model also powers Grok Voice inside Tesla vehicles and other consumer-facing voice features, though xAI has not shipped a standalone transcription app for everyday users the way ElevenLabs or Otter have.
Atlassian’s Loom, which transcribes every video uploaded to the platform, is one of the named early adopters. Atlassian’s Sanchan Saxena was quoted in xAI’s announcement saying the pairing of Grok’s transcription with coding tools “closes the loop from context to code,” a sign xAI is positioning this as developer and workflow infrastructure rather than a consumer product.
Price and Plan Requirements
Pricing did not change from Voice Transcribe 1.0:
- Batch transcription: $0.10 per hour of audio
- Streaming transcription: $0.20 per hour of audio
Diarization, timestamps, and key term detection are bundled in at no additional cost, so teams that were paying extra for those add-ons elsewhere may see a real savings, not just an accuracy bump.
How Users Can Access It
Access runs through xAI’s developer platform and requires an API key, so this is not a one-click consumer tool. Businesses, app developers, and creators who already pipe audio through the Grok API can start using version 2.0 immediately. Anyone who wants to keep using the older model can still pin grok-voice-transcribe-1.0, though xAI says that version will be phased out over the coming weeks.
Limitations
Every accuracy figure in this release, including the “twice as accurate” claim and the 6.8 percent word error rate, comes from xAI’s own internal testing on four production audio sets, not from an independent lab. The Artificial Analysis leaderboard ranking is third-party, but it reflects xAI’s submitted results rather than a fully independent replication. As with any company benchmark, real-world performance on a specific accent, industry vocabulary, or noisy environment can vary from the published numbers.
It is also worth noting this is an API product first. Casual users who just want to transcribe a single video or voice memo will still need a front-end app, browser extension, or existing tool (Loom, for example) that has integrated the Grok API, rather than a dedicated Grok transcription app of their own.
Competitor Comparison
xAI’s internal testing compares Voice Transcribe 2.0 against Google’s Gemini 3.5 Transcribe, Microsoft’s MAI-Transcribe-2, ElevenLabs Scribe v2, Deepgram Nova-3, and OpenAI’s Whisper Large v3, according to reporting from Technobezz. xAI says its model comes out ahead on those internal tests, but none of the other providers have independently confirmed or disputed the comparison, and no neutral third party has re-run the same test set across all six models. Readers comparing transcription tools for their own workflow should treat these as one company’s claims rather than a settled result, and weigh their own accent, language mix, and budget before switching providers. We will keep tracking how it holds up against rivals in our AI tool reviews as more independent testing appears.
What This Means
For AI content creators specifically, the practical upside is cheaper, more accurate auto-captions and multi-speaker transcripts for interviews, tutorials, and reaction videos, especially in mixed-language content. For businesses, bundling diarization and timestamps into the base price removes a cost line that competitors often charge for separately. The bigger picture is that xAI is quietly building out an audio and voice stack (transcription, voice agents, a no-code voice agent builder announced at the same event) rather than only competing on chatbot features, which puts more pressure on Whisper, ElevenLabs, and Deepgram in the transcription market specifically. It is also a reminder that AI infrastructure decisions, not just flashy consumer apps, increasingly shape the AI business tools that companies build on top of them.
Quick FAQs
Is Grok Voice Transcribe 2.0 free? No. It is billed per hour of audio processed, at $0.10 per hour for batch jobs and $0.20 per hour for real-time streaming.
Can I use it without coding? Not directly. It is an API product, so most people will access it through an app or platform (like Loom) that has already integrated it, rather than through a standalone Grok transcription app.
Will my existing Grok transcription integration change automatically? According to xAI, existing API integrations upgrade to version 2.0 automatically, and the older model stays available on request until it is fully phased out.
How does it compare to Whisper or ElevenLabs Scribe? xAI’s own internal tests show it ahead of Whisper Large v3, ElevenLabs Scribe v2, and Deepgram Nova-3, but these are company-reported results, not independent benchmarks, so it is worth testing your own audio before switching.
For more on how AI tools are evolving across chatbots, image, and video generation, keep checking NexPloreInfo’s tech coverage, where we track pricing and feature changes as they happen. We will update this piece as xAI shares independent benchmark results or expands transcription features to more consumer-facing apps.
Leave a Comment