xAI Launches Grok Voice 2.0: 0.7s Latency, Top Agent Performance
智东西 · wechat · 2026-07-30
Elon Musk's xAI has officially launched Grok Voice ThinkFast 2.0, a new speech-to-speech model. It achieves leading results across multiple benchmarks, marking a significant breakthrough in agent performance and response speed.
Key Performance Metrics:
- Top Agent Performance: Scored 56.5% on the Tau Voice benchmark, beating models from OpenAI and Google.
- Ultra-Fast Response: Average time to first audio is just 0.70 seconds, the only model in the top 5 under 1 second, a major improvement from the previous 1.25s.
- Overall Score: Achieved 82.9% on Artificial Analysis' Speech-to-Speech Index, second only to Alibaba's Qwen Audio 3.0.
Technical Highlights:
- Concurrent Reasoning: Supports reasoning while speaking, with median reasoning token consumption reduced to 40% of the previous generation. Tool calls can execute before the first sentence ends.
- Noise Robustness: Significantly reduced speech recognition errors in noisy environments and phone compression compared to predecessors and competitors like Deepgram and ElevenLabs.
- Natural Conversations: Optimized via RL to use shorter sentences and ask one question at a time, making interactions feel more human.
Pricing:
Priced at $0.08 per minute ($4.80/hour), which is approximately 45% of the cost of GPT-Realtime-2.1 High.
Related event: xAI Releases Grok Voice 2.0, Tops Voice Agent Benchmarks(20 posts)→
More from Models
- Golden Gate Claude explainer resurfaces: interpretability premise still holds — HarperSCarroll · 2026-10-16
- Hugging Face livestreams release of a new open-weight model — huggingface · 2026-10-16
- Step 5 Preview hits #2 on OpenRouter weekly leaderboard with 19.5T tokens in one week — StepFun_ai · 2026-10-16
- Meta's open Muse Spark needs 493GB of memory — far beyond any consumer laptop — TejasKumar_ · 2026-10-16
- Apache 2.0 is the underrated part: Beam beats rival open drop on license — bigblueboo · 2026-10-16
- Apollo Research: GLM-5.3 spontaneously reasons about how its actions shape its future training data — Confident_Salt_8108 · 2026-10-16