xAI put Grok STT 1.0 on OpenRouter on July 23, one day before it made noise about it. This is a speech-to-text model — you feed it audio, it hands back text. Nothing exotic in what it does; the fight is over price and accuracy.
And the numbers are loud. $0.10 per hour for batch, $0.20 for streaming. On phone-call entity recognition it posts a 5.0% error rate, against ElevenLabs at 12.0%, Deepgram at 13.5%, and AssemblyAI at 21.3%. If those hold up, xAI just walked into the enterprise voice market and cut the floor out.
What you can build with it
Two ways in: REST for batch jobs, WebSocket for real-time streaming. You get 25 languages, word-level timestamps, speaker diarization, multichannel audio, and inverse text normalization across 12 formats. The obvious use case is the live layer of a voice agent — the ear that feeds a Grok-powered brain and eventually a TTS mouth.
Why it matters
This closes Grok’s voice loop: recognition, reasoning, generation, all callable. And xAI’s real edge is distribution — pipe it straight into the API and X, now OpenRouter too.
You Might Also Like
- Openai gpt Realtime 2 Translate Whisper Three Voice Models one api Several Startups Erased
- Openai gpt Realtime 2 1 gpt Realtime 2 1 Mini cut Voice Agent Latency by 25
- Apple Speechanalyzer api Benchmark 55 Faster Than Whisper and More Accurate in English
- Assemblyais Voice Agent api Undercuts Vapi and Retell With 307ms stt Latency
- Openai Realtime Voice Webrtc Stack the Infra Blueprint Every Voice Agent Startup now has to Compete With
Top AI Product 
Leave a comment