IT之家 AIUpdated

SpaceXAI releases Grok Voice 3.0: billed as the world's smartest voice model, with latency below 1 second

IThome, October 15 news: SpaceXAI today released Grok Voice 3.0, which the company says is the world's most intelligent speech-to-speech model. The model set a new record on customer service call handling tasks, features…

Image source · IT之家 AI
Thanks to IThome netizens South China Wu Yanzu、Yugong Riding a Horse、Buyao Ya for the tip!

IThome October 15 news, SpaceXAI today released Grok Voice 3.0,which the company says is the world's most intelligent speech-to-speech model. The model set a new record on customer service call handling tasks, features an industry-leading latency of under 1 second, and runs at a lower cost than other frontier models. Grok Voice supports millions of conversations daily at Tesla and SpaceX, is now live via an application programming interface, and is priced the same as the previous generation.

In independent testing by Artificial Analysis, Grok Voice 3.0 ranked first on the speech-to-speech overall leaderboard with a score of 86.9. The leaderboard combines voice reasoning ability, the τ-Voice metric, and preference and task completion rates in blind call comparisons. Grok Voice 3.0 scored 97% on the full-duplex benchmark, which measures how naturally the model handles pauses, turn-taking, and interruptions.

τ-Voice evaluates an agent's ability to handle various accents, background noise, and mid-sentence interruptions in customer service call scenarios. Grok Voice 3.0 scored 71.6%, performing on par with GPT-Live-1 (Astra) but with 40% lower latency, opening up a new optimal balance between agent intelligence and response speed.

The company says its running costs are also lower than comparable models. Artificial Analysis tracks the cost of task completion in its Large Audio benchmark, and Grok Voice 3.0 costs less than GPT-Live-1 (Astra, medium version) and GPT-Realtime-2.1.

On ServiceNow's EVA benchmark for enterprise voice agents, Grok Voice 3.0 achieved the highest accuracy score of 74%, and performed best on the combined score of accuracy and customer experience.

According to the company, Grok Voice 3.0 handles millions of conversations daily inside Tesla vehicles. Drivers can use Connectors for simple commands like checking their schedule or replying to email, and use Grok Bot for more complex tasks such as ordering coffee or booking dinner, all hands-free. Grok Voice 3.0 also powers Grok Bot's voice mode.

The Starlink customer service center relies on Grok Voice to handle 20,000 customer calls per day in multiple languages. About 70% of calls are resolved without human intervention, and a quarter of sales calls close a deal before the customer hangs up. A single agent built on Grok Voice 3.0 can verify customer identity, troubleshoot hardware issues, issue refunds, and ship replacement devices.

Grok Voice 3.0 is now live on the voice agent API, priced at $0.08 per minute (IThome note: about 0.54 RMB at current exchange rates) ($4.8 per hour (about 32.2 RMB at current exchange rates)), the same price as Grok Voice 2.0. Agents using the grok-voice-latest version have been upgraded automatically with no changes needed.

Original source

IT之家 AI

Content notes

Original publication and rights belong to the source.

Machine translation · Refer to the original