xAI has released Grok Voice Think Fast 2.0, a new speech-to-speech model announced July 29, 2026. It reasons directly over audio, and xAI says it transcribes 1.5 to 2 times more accurately than dedicated speech-to-text systems on a 24-language evaluation while using roughly 60 percent fewer reasoning tokens than version 1.0. The grok-voice-latest endpoint routes to it automatically starting August 5.

What This Enables

If you are building a voice agent, a single speech-to-speech model removes the usual three-hop stack of transcribe, reason, then synthesize. You send audio and get spoken audio back, with the reasoning handled inline. Fewer reasoning tokens per turn means lower latency and lower cost per conversation, which matters when you are running a live support line, a voice tutor, or an in-app assistant. You can point existing Grok Voice integrations at the new model through the voice API without rebuilding your pipeline.

Why It Matters for Creators

Voice is becoming a first-class build surface, and the transcription accuracy gain is the headline for anyone captioning or indexing audio. A model that both understands and speaks in one pass simplifies the tooling creators need for podcasts, video dubbing, and interactive characters. It lands the same week OpenAI shipped its own transcription models, echoing the wave of native voice modes across every major lab. Independent write-ups from explainX break down the API changes in more depth.

Key Details

Model: grok-voice-think-fast-2.0, a speech-to-speech model.

Accuracy: 1.5 to 2 times better transcription than dedicated STT on a 24-language test, per xAI.

Efficiency: About 60 percent fewer reasoning tokens than Think Fast 1.0.

Pricing: Reported at $0.08 per audio minute, up from $0.05 for version 1.0.

Routing: grok-voice-latest points to this model from August 5, 2026.

What to Do Next

Check the xAI model list to confirm access on your tier, then run a short audio clip through both version 1.0 and 2.0 to measure the latency and accuracy difference before August 5 flips the default. Voice-first products are the clearest place to feel the gain.