Skip to content
Techtember

Always First. Fast Tech News.

A Tesla cabin from the in-car clip on SpaceXSI's Grok Voice 3.0 announcement page
Image: SpaceXSI / Tesla

AI

Grok Voice 3.0 Tops the Artificial Analysis Speech Index and Is Already Talking to Tesla Drivers

SpaceX Super Intelligence says its new speech-to-speech model answers in under a second, costs the same $0.08 a minute as Grok Voice 2.0, and handles millions of conversations a day in Tesla vehicles.

SAN FRANCISCO · Jordan Hale, AI · Oct 15 2026

SAN FRANCISCO - SpaceX Super Intelligence, the Grok developer formerly known as xAI, on Wednesday released Grok Voice 3.0, a speech-to-speech model it calls the most intelligent voice model in the world. It is live now in the company’s Voice Agent API at $0.08 a minute, or $4.80 an hour, the same price as Grok Voice 2.0, and it already powers Grok in Tesla cars and the voice mode in Grok Bot.

“Introducing Grok Voice 3.0, the world’s most intelligent speech-to-speech model,” the company wrote from its @SpaceXSI account on X. “It excels at resolving customer calls, responds in under a second, and costs less than other frontier models.”

Screenshot: @SpaceXSI on X, Oct. 14, 2026

The headline number comes from Artificial Analysis, the independent benchmarking firm. In its testing dated Oct. 13, Grok Voice 3.0 scored 86.9 on the Artificial Analysis Speech-to-Speech Index, ahead of GPT-Live-1 (Astra, Medium) at 83.2, Gemini 3.8 Live at 82.6 and Grok Voice 2.0 at 81.3. The index combines speech reasoning, the tau-Voice support-call test, and preference and task success from blind head-to-head calls. Artificial Analysis confirmed the No. 1 debut in its own post and said the model is also competitive on latency.

Grok Voice 3.0 against Grok Voice 2.0, GPT-Live-1 and Gemini 3.8 Live, using Artificial Analysis data from Oct. 13, 2026. Image: SpaceXSI

The sub-scores are tight at the top and wider underneath. Grok Voice 3.0 posted 98.5 percent on Big Bench Audio for speech reasoning, a 1070 Elo rating in the Speech Agent Arena for human preference, a 94.8 percent task success rate and 97.0 percent on Full Duplex Bench, which measures how naturally a model handles pauses, turn-taking and interruptions. On tau-Voice, which tests support calls with mixed accents, background noise and interruptions, it scored 71.6 percent. SpaceXSI says that is comparable with GPT-Live-1 (Astra) at 40 percent lower latency.

Screenshot: @ArtificialAnlys on X, Oct. 14, 2026

Enterprise testing points the same way. On EVA-Bench, ServiceNow’s open-source benchmark for voice agents, Grok Voice 3.0 reached the highest accuracy score at 74 percent. “Grok Voice 3.0 sets a new Pareto frontier on EVA-Bench,” said Nitin Aggarwal, a senior director of product management at ServiceNow, adding that it “pairs top accuracy with a natural, responsive conversation.”

Grok Voice 3.0 key art. Image: SpaceXSI

The most interesting test bench may be the car. SpaceXSI says Grok Voice 3.0 handles millions of conversations a day in Tesla vehicles. Drivers can use Connectors for simple jobs such as checking a calendar or answering an email, and Grok Bot for bigger tasks such as ordering a coffee or booking dinner, all hands-free. Tesla showed off Connectors in a Sept. 22 post: “@Grok in your Tesla can now do meaningful work for you.”

Screenshot: @Tesla on X, Sept. 22, 2026

Video: Grok with Connectors in a Tesla, @Tesla on X

Starlink is the other proving ground. SpaceXSI says its call centers use Grok Voice to answer 20,000 customer calls a day in multiple languages, that roughly 70 percent are resolved with no human in the loop, and that one in four sales calls ends in a purchase before the caller hangs up. A single agent built on the new model verifies customers, troubleshoots hardware, issues credits and ships replacement equipment.

Another frame from the Tesla in-car clip on the Grok Voice 3.0 page. Image: SpaceXSI / Tesla

For developers, the switch is close to automatic. Agents that call grok-voice-latest have already moved to Grok Voice 3.0 with no code changes, and the company’s release notes now list grok-voice-2.0 as deprecated, with an upgrade path to grok-voice-3.0 at the same price. SpaceXSI also says that on Artificial Analysis’s cost-per-task measure from its Big Bench Audio run, Grok Voice 3.0 comes in under both GPT-Live-1 (Astra, Medium) and GPT-Realtime-2.1. The launch comes days after the company’s X account switched from @SpaceXAI to @SpaceXSI, and it follows Grok Voice Transcribe 2.0 in September.