Issue Nº 01 — October 15, 2026

@readmodelmag
MODEL

Follow MODEL on X @readmodelmag

Tools

Grok Voice 3.0, at the old price

SummaryxAI’s new speech-to-speech model is in the API at eight cents a minute. The ranking it wants quoted comes from an outside benchmark it cites.

· MODEL Editors · 1 min read

A Neumann U87 condenser microphone in a recording studio.
A Neumann U87 in a Richmond, Virginia studio. It is not xAI hardware. Photograph by Will Fisher, via Wikimedia Commons. CC BY-SA 2.0. Image source.

xAI announced Grok Voice 3.0 on October 14. The company calls it a speech-to-speech model with latency under one second, and says it is available in the Voice Agent API at $0.08 a minute, $4.80 an hour, the same price as Grok Voice 2.0. Agents already pointed at grok-voice-latest were switched over. No integration change, xAI says.

The benchmark xAI leads with is Artificial Analysis’s Speech-to-Speech Index. xAI dates that source October 13, 2026, and reports a score of 86.9 for Grok Voice 3.0, against 83.2 for GPT-Live-1 at its Astra medium setting and 82.6 for Gemini 3.8 Live with extended thinking set high. On the same table, xAI lists Full Duplex Bench at 97.0 percent, a task-success rate of 94.8 percent, and a speech-reasoning score, on Big Bench Audio, of 98.5 percent. On τ-Voice, the support-call test with accents, noise, and interruptions, xAI reports 71.6 percent and says that is comparable to GPT-Live-1 at 40 percent lower latency.

MODEL has not replayed those calls. The numbers are xAI’s citation of Artificial Analysis. If you are buying voice for a support line, the table is a reason to run your own calls.

Two deployment figures, again xAI’s. The company says Grok Voice handles millions of conversations a day in Tesla vehicles, including connectors for calendar and email and a mode it calls Grok Bot for tasks such as ordering or booking. It says Starlink call centers use Grok Voice for 20,000 customer calls a day, across multiple languages; that roughly 70 percent are resolved with no person in the loop; and that one in four sales calls ends in a purchase before the caller hangs up. A single agent, xAI says, verifies customers, troubleshoots hardware, issues credits, and ships replacements.

On ServiceNow’s EVA-Bench, xAI reports the highest accuracy score at 74 percent. That figure is xAI’s account of the benchmark.

The price is the cleanest fact on the page. Eight cents a minute, same as last time, live in the API now.

Sources

  1. Introducing Grok Voice 3.0

    xAI

The letter

Friday, three tools.

Three tools. Tested. The rest is noise.

The form posts to the Beehiiv subscribe page. The weekly three is editorial. A subscription does not buy a place in it.

  • Programming source code on a computer monitor.

    The fitting

    Graphicle, on the stand

    A free local Rust binary draws a code graph for a coding agent. We ran the Linux build. The speed claim belongs to its maker.

    · MODEL Editors

  • The entrance sign at Meta Platforms headquarters in Menlo Park, California.

    Tools

    Meta sends out Muse Wonder

    Two agent models, a published API price, and a subscription tier with eighty times the ordinary limit. The benchmark glory is Meta’s to defend.

    · MODEL Editors

  • TSMC’s Fab 5 building in the Hsinchu Science Park.

    Markets

    TSMC’s third quarter, in its own numbers

    The foundry reported NT$1,494.24 billion of revenue and NT$29.43 a share for the quarter ended September 30. The fourth-quarter range that follows is management’s outlook, not ours.

    · MODEL Editors