SpaceXAI's Grok Voice 3.0 tops the speech index at $0.08 a minute
Voice agents are having their price-war moment: a new frontier speech model landed today at the same per-minute rate as the one it replaces, and it is already answering the phones at Starlink. Meanwhile the AI buildout keeps pulling power plants out of the ground, and a license-plate camera lawsuit tests whether surveillance guardrails mean anything.
SpaceXAI released Grok Voice 3.0, a speech-to-speech model it says ranks first on Artificial Analysis' independent Speech-to-Speech Index with a score of 86.9 — ahead of OpenAI's GPT-Live-1 (Astra, Medium) at 83.2 and Google's Gemini 3.8 Live Extended Thinking at 82.6. The index blends speech reasoning, the τ-Voice customer-support benchmark, arena preference from blind head-to-head calls, and task success; Grok Voice 3.0 also scores 97% on Full Duplex Bench, which measures how naturally a model handles pauses, turn-taking, and interruptions. It is not a clean sweep. On τ-Voice itself — simulated support calls with noisy audio, accents, and interruptions — Grok Voice 3.0 lands at 71.6%, behind GPT-Live-1 Astra's 74.5%, and SpaceXAI's pitch is the trade rather than the win: comparable accuracy at roughly 40% lower latency, with time to first audio of 0.81 seconds. On Artificial Analysis' broader field it is not the fastest model either — Raon SpeechChat answers in 0.04 seconds — it is the fastest one that also sits near the top of the quality tables.
The deployment numbers are the more interesting half. SpaceXAI says Grok Voice answers 20,000 Starlink customer calls a day across multiple languages, resolves roughly 70% 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 model verifies customers, troubleshoots hardware, issues credits, and ships replacement gear. It also runs in Tesla vehicles, and on ServiceNow's EVA-Bench for enterprise voice agents it scores the highest accuracy at 74%. Crucially, the price did not move: $0.08 per minute, unchanged from Grok Voice 2.0, and agents pointed at the latest model were migrated automatically.
Our read: the flat rate is the story. Two months ago Google launched a voice model at a sixth of GPT-Live's price and framed it as commoditizing the layer; today the incumbent-quality model holds its old rate while adding capability, which is the same squeeze from the other direction. The counterweight is that "frontier voice" now means resolving about 70% of real support calls, not all of them — the last third is where the margin and the headcount still live. We covered the last round of voice-benchmark claims in September — Gemini 3.8 Live tops voice benchmarks at a sixth of GPT-Live's price.
Ambrosia Energy is building a 10 gigawatt-hour battery factory near Austin, Texas, to supply the off-grid power plants it sells to data centers. Co-founder and CEO Ben Longmier announced the 150,000-square-foot plant on stage at TechCrunch Disrupt, saying it lets the company build roughly a 300-megawatt power plant per year at its field sites, with full production in two to three years. The startup emerged from stealth in June with a $100-per-megawatt-hour target for solar-plus-battery plants — the pair of former SpaceX engineers behind it previously founded the satellite startup Swarm — and it is starting its first commercial plant in West Texas later this year: 300 megawatts of solar paired with 660 megawatt-hours of batteries to deliver 30 megawatts of continuous power, aiming for 99.3% uptime. Sections of the pilot plant have run continuously for more than 100 days at 200 kilowatts. Why it matters: the binding constraint on AI capacity has moved from chips to electrons, and $100 per megawatt-hour puts batteries in direct competition with the natural-gas turbines most hyperscalers are planning.
A Georgia small-business owner is suing Flock Safety and a county sheriff's office, alleging deputies used the company's license-plate cameras to track him without probable cause after he sued the department over excessive force. Tyler Murkerson's complaint, filed October 8, says his plate was searched repeatedly while his earlier excessive-force case was pending, and that the only justifications logged were "info" and "37" — neither tied to any case, incident, or call for service. It is the weakness critics have flagged for years: Flock's platform accepted free text in the justification field, and one department logged "hehehe" twenty separate times. Flock said in August it would require case codes for all law-enforcement searches by the end of the year, and Coweta County began its own audit that month — three employees resigned before it finished. Hundreds of agencies have dropped Flock contracts this year. The company says it stands by its technology and will defend the claims in court.
What to watch: whether Flock's case-code requirement actually lands by year end, and whether any other agency gets a lawsuit like this one.
Is a guardrail that arrives after the abuse a guardrail at all — or just paperwork? Tell us in the comments.
