%20--%3e%3cdefs%3e%3cstyle%3e%20.st0%20{%20fill:%20%23061b40;%20}%20.st1%20{%20fill:%20%23306af1;%20}%20.st2%20{%20fill:%20%235ce5cf;%20}%20%3c/style%3e%3c/defs%3e%3cg%3e%3cpath%20class='st0'%20d='M55,10.5h9v3h-9v9h12V7.5h-12v3ZM64,19.5h-6v-3h6v3Z'/%3e%3cpolygon%20class='st0'%20points='69%2016.5%2078%2016.5%2078%2019.5%2069%2019.5%2069%2022.5%2081%2022.5%2081%2013.5%2072%2013.5%2072%2010.5%2081%2010.5%2081%207.5%2069%207.5%2069%2016.5'/%3e%3cpolygon%20class='st0'%20points='95%2010.5%2095%207.5%2083%207.5%2083%2022.5%2095%2022.5%2095%2019.5%2086%2019.5%2086%2016.5%2095%2016.5%2095%2013.5%2086%2013.5%2086%2010.5%2095%2010.5'/%3e%3cpath%20class='st0'%20d='M40,1.5v21h11.6l1.4-1.4v-7.6h0c0,0-1.4-1.5-1.4-1.5l1.4-1.4V2.9l-1.4-1.4h-11.6ZM50,19.5h-7v-6h7v6ZM50,10.5h-7v-6h7v6Z'/%3e%3c/g%3e%3cpath%20class='st1'%20d='M23.1,24L14.7,4.8l-4.9,11.2h3.8l-1.8,4H3.3L12.1,0H2C.9,0,0,.9,0,2v20c0,1.1.9,2,2,2h21.1Z'/%3e%3cpath%20class='st2'%20d='M34,0h-16.8l10.6,24h6.2c1.1,0,2-.9,2-2V2C36,.9,35.1,0,34,0ZM32.5,20h-4V4h4v16Z'/%3e%3c/svg%3e)
xAI has launched its new speech recognition model, Grok Voice Think Fast2.0, which is now available to developers building voice-powered agents. It offers improvements in intelligence level, transcription accuracy, dialogue capabilities, and tool calling efficiency. The model is priced at $0.08 per minute of audio. xAI states that no changes are needed to existing prompts, and Think Fast2.0 can bring performance improvements in most application scenarios.
In the Artificial Analysis speech recognition benchmark test, Grok Voice Think Fast2.0 achieved a comprehensive score of 82.9%, higher than the previous version Think Fast1.0's 75.7%, as well as exceeding GPT-Realtime-2.1's 79.1% and Gemini3.1Flash's 69.5%. Its agent capability score reached 56.5%, leading Think Fast1.0's 52.1%, GPT-Realtime-2.1's 45.7%, and Gemini3.1Flash's 37.7%. The model's first audio playback time was reduced from 1.25 seconds to 0.70 seconds, further improving response speed.
In terms of speech transcription, xAI tested large amounts of voice clips in 24 languages. Official data shows that compared to Deepgram Nova3 and ElevenLabs Scribe v2, the transcription accuracy of Think Fast2.0 has improved by about 1.5 to 2 times; compared to the previous generation model, it has improved by about 1.4 times. In complex environments such as background noise and phone compression, xAI says the difference can expand to about 10 times.
On the technical side, Think Fast2.0 supports parallel speech inference, improving processing efficiency for complex tasks by reducing waiting time. Compared to the previous generation model, the median number of inference tokens used has dropped to 0.4, allowing tool calls to typically be executed before the agent completes its first sentence reply. At the same time, reinforcement learning has optimized the model's dialogue strategy, making it tend to use shorter answers, handle one issue at a time, and reduce the output of irrelevant information, thereby better supporting complex workflows.
xAI stated that this upgrade mainly focuses on enhancing the reliability of voice Agents in real business scenarios. Currently, Grok Voice is undergoing A/B testing in Starlink telephone services and has brought improvements in sales conversion rates and customer service response efficiency.
According to the plan, on August 5, 2026, the grok-voice-latest alias will automatically switch from grok-voice-think-fast-1.0 to grok-voice-think-fast-2.0. Developers who still need to use the old version should lock the 1.0 version identifier in advance. Other users do not need any additional actions.
As AI Agents gradually enter practical application scenarios such as customer service, sales, and office work, speech models with low latency, high accuracy, and multi-tool collaboration are becoming an important foundation for the next stage of intelligent interaction.