xAI announced Grok Voice 3.0 on 14 October 2026. It said the model is available today in its Voice Agent API at $0.08 a minute, or $4.80 an hour, the same price as Grok Voice 2.0. xAI says that in Artificial Analysis’s testing the model ranks first on the Speech-to-Speech Index with a score of 86.9, against 83.2 for OpenAI’s GPT-Live-1 (Astra, Medium).
The SpaceX Super Intelligence account on X posted the launch the same day. xAI says Grok Voice handles millions of conversations a day in Tesla vehicles, and that Starlink call centers use it to answer 20,000 customer calls a day, with roughly 70% resolved with no human in the loop.
Business Pill · COST PER TASK
A one-minute explainer of cost per task: paying for the job done, not the word or the minute. It teaches the general idea only and says nothing about any organisation in this story.
The key insight: As we read it, xAI is selling Grok Voice 3.0 on cost per completed call rather than on price per minute: the minute rate is unchanged at $0.08, and the claims xAI leads with are task success, resolution without a human and cost per task against OpenAI’s models.
xAI calls Grok Voice 3.0 “the world’s most intelligent speech-to-speech model”. It says the model “sets a new record for resolving customer support calls, has class-leading latency at under one second, and costs less than other frontier models.”
The page says the model is available in the Voice Agent API and that agents using grok-voice-latest have already been switched over automatically, with no changes needed. The SpaceX Super Intelligence posts add that it is available in the API and Voice Agent Builder at $0.08 a minute.
Artificial Analysis Speech-to-Speech Index scores as xAI reports them on 14 October 2026: Grok Voice 3.0 at 86.9, OpenAI’s GPT-Live-1 (Astra, Medium) at 83.2, Gemini 3.8 Live at 82.6 and Grok Voice 2.0 at 81.3. Higher is better.
What Was Posted on X
Introducing Grok Voice 3.0, the worldโs most intelligent speech-to-speech model.
It excels at resolving customer calls, responds in under a second, and costs less than other frontier models. pic.twitter.com/cfDRrkQPEc
The same account’s follow-up giving the price this piece leads with: $0.08 a minute in the API and Voice Agent Builder.
SpaceX has just introduced Grok Voice 3.0
"It's the worldโs most intelligent speech-to-speech model. It sets a new record for resolving customer support calls, has class-leading latency at under one second, and costs less than other frontier models. It's also cheaper to run thanโฆ pic.twitter.com/Ll3aFALE3p
Sawyer Merritt’s post quoting xAI’s launch text, including the cost-per-task comparison with GPT-Live-1 and GPT-Realtime-2.1.
The Price and the Cost per Task
xAI states the price as $0.08 a minute ($4.80 an hour), unchanged from Grok Voice 2.0. By our arithmetic, a ten-minute support call costs $0.80 in API fees at that rate.
On running cost, xAI says: “Artificial Analysis measures what each model costs to complete a task in its Big Bench Audio run, and Grok Voice 3.0 comes in under both GPT-Live-1 (Astra, Medium) and GPT-Realtime-2.1.” The page shows those per-task costs in a chart; we do not quote dollar figures from it.
The Benchmarks xAI Cites
xAI’s table, which it attributes to Artificial Analysis, puts Grok Voice 3.0 at 98.5% on Big Bench Audio speech reasoning, against 90.1% for GPT-Live-1 and 97.7% for Gemini 3.8 Live (Extended Thinking, High).
The same table gives a task success rate of 94.8% for Grok Voice 3.0, against 87.4% for GPT-Live-1 and 89.1% for Gemini 3.8 Live, and a Speech Agent Arena Elo of 1070, against 1048 and 990.
On ฯ-Voice, which xAI describes as testing support calls with diverse accents, background noise and interruptions, xAI says the model scores 71.6%, comparable with GPT-Live-1 (Astra), “but at 40% lower latency.”
xAI also says the model reaches the highest accuracy score, 74%, on ServiceNow’s EVA-Bench for enterprise voice agents. It quotes Nitin Aggarwal of ServiceNow: “Grok Voice 3.0 sets a new Pareto frontier on EVA-Bench”.
What xAI Says About Starlink and Tesla
xAI says Starlink call centers use Grok Voice to answer 20,000 customer calls every day across multiple languages. It says roughly 70% are resolved with no human in the loop, and that one in four sales calls ends in a purchase before the caller hangs up.
It says a single agent built on Grok Voice 3.0 verifies customers, troubleshoots hardware, issues credits and ships replacement equipment.
In Tesla vehicles, xAI says, drivers use Connectors for requests such as checking a calendar or answering an email, and Grok Bot for tasks such as ordering a coffee or booking dinner. It says the model also powers Grok Bot’s voice mode. On 13 October we covered how SpaceX bundled Grok, Cursor and X Premium into one subscription.
The Structural Read
As we read it, holding the price flat while the benchmark scores rise is a price cut per unit of work, if xAI’s task-success figures hold outside the test.
As we read it, the Starlink figures are the commercial argument: xAI says roughly 70% of 20,000 daily calls end with no human in the loop, which is the number a support buyer prices against staff cost.
As we read it, xAI says Tesla vehicles and Starlink call centres already run Grok Voice at scale, which is the distribution point of the launch.
Nitin Aggarwal, Senior Director of Product Management, Agentic Evals and RAI, ServiceNow, as quoted by xAI, 14 October 2026
“It pairs top accuracy with a natural, responsive conversation, setting a new standard for what enterprises can expect from AI on the front line of customer service”
Three Implications
SUPPORT AND CONTACT-CENTRE BUYERS xAI says a single Grok Voice 3.0 agent verifies customers, troubleshoots hardware, issues credits and ships replacement equipment at Starlink.
VOICE DEVELOPERS ON THE API xAI says agents on grok-voice-latest were switched over automatically, at the same $0.08 a minute as Grok Voice 2.0.
OPENAI’S REALTIME VOICE LINE xAI says Artificial Analysis’s cost per task puts Grok Voice 3.0 under both GPT-Live-1 (Astra, Medium) and GPT-Realtime-2.1.
The framework puts it this way: “The new job of finance is to manage the ratio between the two — work produced per token burned. That ratio is the gross-margin equation of the agentic era.”
As we read it, xAI’s pitch is written in that ratio: the same price a minute, with more calls resolved per minute bought, by its own account.
xAI’s stated price for Grok Voice 3.0 in its Voice Agent API: $0.08 a minute ($4.80 an hour), the same as Grok Voice 2.0. The Starlink and latency figures are xAI’s own.
What Is Not Established
The benchmark scores and cost comparison are xAI’s report of Artificial Analysis’s testing; we read xAI’s page, not Artificial Analysis’s own results pages. The Starlink and Tesla figures are xAI’s own and come with no outside data.
xAI gives the cost-per-task comparison as a chart without figures in its text, so the size of the gap is not stated in the material we read. OpenAI had made no comment on the comparison in the material we read.
Business Engineer Framework
The CFO’s Guide to the Token Economy
A Business Engineer framework on why cheaper tokens can still mean bigger AI bills, and how to measure work produced per token.
xAI launched Grok Voice 3.0 on 14 October 2026 at $0.08 a minute in its Voice Agent API, the same price as Grok Voice 2.0, and says Artificial Analysis’s testing ranks it first on the Speech-to-Speech Index at 86.9, ahead of OpenAI’s GPT-Live-1 at 83.2.
A note on sourcing. On 14 October 2026 we read xAI’s page introducing Grok Voice 3.0 in full, the launch posts by the SpaceX Super Intelligence account on X, and the post by Sawyer Merritt that drew our attention to it. The benchmark scores and the cost comparison are xAI’s report of Artificial Analysis’s testing; we did not read Artificial Analysis’s own results. The per-call arithmetic is ours, at the price xAI states.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur, HackerNoon and Search Engine People. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
