xAI has opened standalone Grok speech-to-text (STT) and text-to-speech (TTS) APIs to outside developers. The company said the release is based on the same production voice stack already running in Grok Voice, Tesla vehicle voice interactions, and Starlink customer support, rather than a separate experimental system.
STT comes in batch and streaming modes
According to xAI, the STT API can be accessed through a REST API for batch transcription and a WebSocket API for low-latency streaming. Pricing is set at $0.10 per hour for batch processing and $0.20 per hour for streaming. xAI said the pricing is lower than major competitors such as ElevenLabs and Deepgram.
Grok STT supports more than 25 languages and includes word-level timestamps, speaker diarization, multi-channel audio support, and smart inverse text normalization. These features target enterprise use cases where transcription accuracy matters, including meetings, legal records, medical documentation, and customer service call logs. The product focus is clear.
Benchmark data highlights named-entity recognition
In a benchmark focused on identifying key entities in phone calls, such as names, account numbers, and dates, xAI said Grok STT posted an error rate of 5.0%. That compares with 12.0% for ElevenLabs, 13.5% for Deepgram, and 21.3% for AssemblyAI. Those figures stand out as one of the central claims in the rollout.
TTS adds five voice styles and inline speech controls
On the synthesis side, the Grok TTS API includes five voice options: Ara, Eve, Leo, Rex, and Sal, spanning female, male, and neutral styles. The service automatically detects the input language, supports more than 20 languages natively, and uses BCP-47 language codes for pronunciation control.
Output formats include MP3, WAV, PCM (Linear16), G.711 μ-law, and G.711 A-law. The last two are common telephony codecs, pointing to compatibility with phone-based systems. xAI priced the TTS service at $4.20 per million characters.
The TTS API also includes “speech tags,” allowing developers to embed instructions in text to control pauses, laughter, whispering, emphasis, speaking rate, and pitch. That gives developers tighter control over how synthetic speech is delivered.
The same infrastructure is already deployed in xAI products
xAI said the newly released STT and TTS endpoints come from the same infrastructure already used across its own products. The company linked the launch back to the Grok Voice Agent API introduced in late 2025, which was positioned as a real-time conversational voice agent stack.
At that time, xAI said the system ranked first on the Big Bench Audio benchmark and delivered a first audio response in under 1 second, roughly 5 times faster than the nearest competitor. By splitting that integrated voice pipeline into separate API endpoints, xAI is now letting developers pick individual components or assemble their own end-to-end voice applications.