VercelEngineering PostOfficial Source

Grok Voice Think Fast 2.0 now available on AI Gateway

Grok Voice Think Fast 2.0 brings speech-to-speech reasoning and earlier tool calls to Vercel’s realtime API, with server-minted tokens keeping gateway keys off clients.

Vercel · Jul 29, 2026

Open Source Open Markdown Open JSON

Source Summary

**Grok Voice Think Fast 2.0** is available through AI Gateway as an audio-in, audio-out model. Vercel says it reasons while speaking, uses fewer reasoning tokens than its predecessor, and handles background noise and telephony compression.

Practical Implication

Voice-agent builders can call **xai/grok-voice-think-fast-2.0** through the AI SDK realtime API. Mint a short-lived token on the server so the gateway API key never reaches browser or app clients.

Agent-Ready Context

**Grok Voice Think Fast 2.0** is available through AI Gateway as an audio-in, audio-out model. Vercel says it reasons while speaking, uses fewer reasoning tokens than its predecessor, and handles background noise and telephony compression.

Voice-agent builders can call **xai/grok-voice-think-fast-2.0** through the AI SDK realtime API. Mint a short-lived token on the server so the gateway API key never reaches browser or app clients.

The claimed improvements in reasoning, transcription, and conversation come without benchmark figures in the material. Earlier tool calls are described as common, not guaranteed, so test interruption timing and noisy-audio behavior on your own workload.

Connected Context · Feed7 Judgment

This extends the candidate model-selection landscape into realtime, bidirectional voice, where latency, interruption timing, noisy input, and client credential exposure matter alongside reasoning quality. The claims do not justify replacing existing routes without workload tests: early tool calls are only typical, improvements lack benchmark figures, and short-lived server-minted tokens are a deployment prerequisite.

X$^3$-OPD: Distilling Reasoning into Large Audio-Language Models via On-Policy AlignmentX³-OPD supplies reinforcing research context that audio reasoning depends on grounding in acoustic events, prosody, and dialogue—the kinds of conditions the voice model’s noise claims should be tested against.Grok 4.5 now available on AI GatewayGrok 4.5 offers adjustable text-and-image reasoning, while Think Fast 2.0 adds a realtime audio route whose selection must account for conversational latency and interruption behavior.Inkling Small from Thinking Machines is now available on AI GatewayBoth expose efficiency-oriented model choices through AI Gateway, but the voice model shifts evaluation from adjustable thinking effort to end-to-end audio quality and realtime behavior.Reasoning LLM Improves Speaker Recognition in Long-form TV DramasThe speaker-recognition result shows a separate audio workload where reasoning and multimodal tools outperform acoustic baselines, reinforcing that audio agents should be evaluated on task outcomes rather than transcription claims alone.

Context Map

modelaudio#reasoning#tool-use#generative-media

Uncertainty

The claimed improvements in reasoning, transcription, and conversation come without benchmark figures in the material. Earlier tool calls are described as common, not guaranteed, so test interruption timing and noisy-audio behavior on your own workload.