xAI just moved into voice infrastructure. The company launched a standalone Grok Speech to Text API, turning audio into text for developers who want transcription without building their own model. According to xAI, the API runs on the same audio stack that already powers Grok Voice, Tesla vehicles, and Starlink customer support, so this isn’t a science project. It’s production tech being opened up for outside use.
What stands out here is the price. xAI is charging $0.10 per hour for batch transcription and $0.20 per hour for real-time streaming. That’s aggressive, and it’s clearly meant to undercut the established players.
🎙️ What the API actually does
Here’s the breakdown of what xAI shipped:
- Two ways to transcribe. A REST endpoint handles file-based batch jobs (think uploading a recorded meeting or podcast), while a streaming endpoint delivers real-time, low-latency transcription for live use.
- 25+ languages. The model covers a wide language range, which matters for anyone building products beyond an English-only market.
- Speaker diarization. It doesn’t just transcribe words, it labels who said what. That’s the difference between a wall of text and a usable meeting transcript.
- Word-level timestamps. Each word gets a time marker, which makes searching, captioning, and syncing audio to text far easier.
- Low latency for live work. xAI is pitching this for voice commands, live captions, and support calls where a delay kills the experience.
💰 How it compares
The pricing is the headline. xAI positions Speech to Text as the best price in the market, and the numbers back the claim up. Word-level timestamps plus speaker diarization usually land in the enterprise tier, where you pay a premium. xAI is bundling those features at a rate that undercuts most competitors.
For context, xAI released this alongside a Text to Speech API priced at $4.20 per million characters, rounding out a full voice toolkit. The transcription market already has serious names in it, including OpenAI’s Whisper-based offerings and ElevenLabs, which rolled out its own speech-to-text model last year. xAI’s play is simple: match the enterprise feature set, then win on cost.
🛠️ Who this is for
The target is developers building voice into their products. A few obvious use cases from the launch:
- Meeting and call transcription, with speaker labels baked in.
- Podcast and media captioning at scale.
- Voice command and voice assistant features that need real-time response.
- Customer support tooling, the exact scenario xAI already runs internally for Starlink.
The fact that this same infrastructure handles Tesla and Starlink workloads is a real signal. It’s been battle-tested on scale before it hit the public API.
⚠️ What to watch
A couple of caveats worth keeping in mind. xAI’s own positioning frames the accuracy and language coverage, but independent benchmarks against Whisper and ElevenLabs will tell the real story once developers put it through its paces. Cheap transcription that misses words costs more than it saves. The 25+ language count is solid, though it’s worth checking whether your specific language gets full accuracy or just basic support.
This is significant because it marks xAI’s shift from a chatbot company into a full-stack AI infrastructure provider. Grok started as an answer to ChatGPT. Now xAI is selling the plumbing underneath it, the same move OpenAI and Google made when they opened their models to developers. Voice is one of the stickiest layers to own, because once a product’s transcription runs on your API, switching costs get real.
The aggressive pricing tells you xAI wants market share fast, not fat margins. Expect the incumbents to respond. When a well-funded player prices this low, everyone else has to explain why they cost more.
Developers can start building against the API now. Full technical details and endpoint documentation are available at the original source.
Sources: xAI news: Grok STT and TTS APIs, xAI Docs: Speech to Text, MarkTechPost