Grok Voice 3.0 Pricing: $0.08 a Minute and What 1,000 Calls Cost

October 15, 2026. xAI released Grok Voice 3.0 on October 14 and kept the price at $0.08 per minute of audio, or $4.80 an hour, in its Voice Agent API, the same rate it charges for Grok Voice 2.0, according to the Grok Voice 3.0 announcement and xAI's API pricing page. That rate is 60 percent above the $0.05 a minute xAI listed in early July, before Grok Voice 2.0 shipped. For a business that answers 1,000 calls a month at three minutes each, the model alone now costs $240 a month, against $150 at the July rate, before you pay for the phone line.

What xAI released on October 14

1. Grok Voice 3.0 in the API. The model ID is grok-voice-3.0, and xAI's release notes say the grok-voice-latest alias now points to it, so agents that use the alias were upgraded automatically. The speech to speech docs say the endpoint, session parameters and events are unchanged.
2. Grok Voice 2.0 is deprecated. Agents pinned to grok-voice-2.0 should switch the model string to grok-voice-3.0 or grok-voice-latest. Both models are listed at the same $0.08 a minute.
3. A benchmark lead. On the Speech to Speech Index from Artificial Analysis, which we read on October 15, Grok Voice 3.0 scores 86.9, ahead of OpenAI's GPT-Live-1 (Astra, medium) at 83.2, Google's Gemini 3.8 Live with extended thinking at 82.6 and Grok Voice 2.0 at 81.3. Its score on τ-Voice, a test of support calls with accents, background noise and interruptions, rose from 56.5 to 71.6 percent.
4. Production claims from xAI. xAI says Starlink call centers answer 20,000 calls a day on Grok Voice, resolve roughly 70 percent with no human involved, and close a purchase on one in four sales calls. It also cites a 74 percent accuracy score on ServiceNow's EVA-Bench. These are vendor figures, not independent tests.

The Neuron's October 14 digest also reported the release and the unchanged API price.

What Grok Voice costs, line by line

xAI bills a voice agent on one meter instead of separate charges for transcription, the model and the voice. Its Voice Agent Builder post says voices are included and there is no separate platform fee. These are the lines that can appear on the bill:

1. Speech to speech, $0.08 a minute. $4.80 an hour of audio for Grok Voice 3.0 or 2.0.
2. Text inputs, $0.004 each. The pricing page lists this next to the per-minute rate for both speech to speech models.
3. Telephony, $0.01 a minute or your carrier's rate. An agent built in the Voice Agent Builder can use a free provisioned number at $0.01 a minute. Through the API, xAI's SIP guide says it does not provision numbers, so you register your own number on your own SIP trunk and pay your carrier. Twilio's US SIP trunking prices are $0.0034 a minute for inbound calls to a local number and $0.0130 to a toll-free number, plus $1.15 or $2.15 a month for the number.
4. The standalone speech APIs. If you build your own pipeline instead, speech to text costs $0.10 an hour on REST or $0.20 streaming, and text to speech costs $15 per million characters.

The price rose in July, not this week

The 60 percent increase did not arrive with 3.0. A capture of the Voice Agent Builder post saved by the Internet Archive on July 4, 2026 lists the API rate as $0.05 a minute of audio. When xAI launched Grok Voice 2.0 on July 29, it priced the model at $0.08 a minute and set the grok-voice-latest alias to switch to 2.0 on August 5. The same builder post now says $0.08. Artificial Analysis lists Grok Voice 1.0 at $3.00 per hour of input audio and both later versions at $4.80, which matches the two rates.

What the extra 3 cents a minute buys, on Artificial Analysis's numbers: an index score of 86.9 against 73.7 for Grok Voice 1.0, a task success rate of 94.8 percent against 80.7, and 0.81 seconds to first audio against 1.25. One trade-off is worth testing on your own calls: Grok Voice 2.0 was quicker to first audio, at 0.70 seconds.

What a 1,000-call month costs

We priced an inbound receptionist that answers 1,000 calls a month at an average of three minutes, so 3,000 minutes. The call length is our assumption, so swap in your own.

1. Model only: $240 a month. 3,000 minutes at $0.08. At the early July rate of $0.05 the same line was $150, so the month costs $90 more than a July budget assumed.
2. On the builder's free number: $270 a month. Telephony adds $30 at $0.01 a minute. That is $0.09 a minute, or about $0.27 for a three-minute call.
3. On your own Twilio SIP trunk: $251.35 a month. $240 plus $10.20 of inbound minutes and $1.15 for a local number, before taxes and carrier fees. A toll-free number takes the month to $281.15.

For the same workload on hosted voice agent platforms, see our Retell AI pricing and Vapi pricing breakdowns and the AI receptionist cost comparison.

How Grok Voice compares with OpenAI and Google

Vendors bill voice models differently, by the minute or by audio tokens, so a like for like price is hard to read off a rate card. Artificial Analysis measures what each model costs to run the same fixed 40-question Big Bench Audio set, normalized to an hour of input audio:

1. GPT-Realtime-2.1 (High): $10.75 an hour. It takes 1.21 seconds to first audio.
2. GPT-Live-1 (Astra, medium): $5.83 an hour. It scores higher on τ-Voice, 74.5 percent against 71.6, but takes 1.34 seconds to first audio against Grok Voice 3.0's 0.81.
3. Grok Voice 3.0: $4.80 an hour. The same as its list price of $0.08 a minute.
4. Gemini 3.8 Live, extended thinking (High): $3.50 an hour. It takes 1.35 seconds to first audio and scores 68.6 percent on τ-Voice.

On that measure Grok Voice 3.0 costs less than both OpenAI models and more than Google's, and it reaches first audio faster than all three. Your own bill depends on call length and on how much of each call the caller spends talking, so run a week of real calls before you move a client's line.

What it means for operators

1. Check which model your agents run. Agents on grok-voice-latest moved to 3.0 on October 14 without a deploy. Replay a set of recorded test calls this week, because timing and tool use can shift with a new model.
2. Re-budget at $0.08. Any quote or client budget built on xAI's early July rate of $0.05 needs its model line raised by 60 percent.
3. Choose the phone line on purpose. The builder's number is the simplest route. Your own SIP trunk is cheaper on inbound local calls, and the number stays with your carrier if you change models later.
4. Outbound calls still need consent. Consent and disclosure rules for automated calls apply whichever model is speaking, so read our TCPA-compliant AI calling page and the AI caller disclosure laws by state before you dial.

In home services the first job for a voice agent is usually the after-hours line. Our AI call answering for home services puts an agent on HVAC, plumbing and roofing phone lines, including after-hours emergencies, and our AI answering service for HVAC comparison shows what the off-the-shelf alternatives cost. If you want a voice agent built on Grok, OpenAI or Gemini and tested on your own calls, see our AI voice agent service.