Grok's voice feature hits number one, plus the AI product news you actually use — in about 15 minutes.
Alex: Good morning, and welcome to the Charles' Daily AI Tech Info by MUSE, for Thursday, October fifteenth. I'm Alex.
Jordan: And I'm Jordan. Big day yesterday in the world of product AI. Grok now has the number one voice model in the world, at least according to the numbers.
Alex: Meta's Muse is being called a bigger threat to Google Search than ChatGPT ever was.
Jordan: Amazon killed the Fire tablet and launched something smarter in its place. And Google is about to read your text messages, but in a good way.
Alex: That's the promise, anyway. Fifteen minutes, everything you need. Let's get into it.
Alex: Headline number one, and it's a Grok story. Yesterday, the fourteenth, xAI announced Grok Voice three point oh, and it is now officially the top-ranked voice AI model in the world.
Jordan: Top-ranked according to who?
Alex: According to Artificial Analysis, the independent benchmarking outfit that the whole AI industry actually pays attention to. They keep a speech-to-speech index that blends four different tests: how well a model reasons through speech, how well it handles voice agent tasks, how real people rank it in live arena testing, and its overall task success rate. Grok Voice three point oh scored eighty-six point nine.
Jordan: And the competition?
Alex: Grok Voice two point oh, the old one, was at eighty-one point three. OpenAI's voice model, which runs on G P T six Astra, came in at eighty-three point two. And Google's Gemini three point eight Live landed at eighty-two point six. So Grok didn't just win. It won by a real margin.
Jordan: Okay, but what does "best voice AI" mean for a normal person? I'm not building an app.
Alex: Fair. This launched in xAI's voice-agent API, which is the developer tool that lets apps and services carry on real-time spoken conversations. It responds to your first word in under a second, point eight one seconds to be exact. It can clone a custom voice from just a two-minute reference clip. And it scored ninety-seven percent on full duplex testing, which means it can handle interruptions and back-and-forth like an actual conversation, not the old wait-for-the-beep voice bots.
Jordan: So every app with a voice feature just got smarter?
Alex: In theory, yes. And here's the number that convinced me this is real: ServiceNow, the big enterprise software company, runs its own test for complete multi-turn spoken conversations, and Grok Voice three point oh hit seventy-four percent accuracy there. The bar for voice AI genuinely went up yesterday.
Jordan: What's this going to cost developers who want to use it?
Alex: xAI lists it at eight cents a minute. For comparison, OpenAI's voice API runs about five cents a minute plus the cost of whatever model is running behind it, and Google's Gemini Live audio is way cheaper per minute but you pay for every text token on top. So Grok isn't the cheapest, but if it's the best-sounding and the most reliable at actually completing tasks, eight cents a minute is nothing for a business running customer calls.
Jordan: All right, I'll give it to you. That does sound legitimately impressive.
Alex: It is. And by the way, this was NOT sponsored in any way. This is an honest review. Grok just keeps shipping, and this time the independent numbers back it up.
Jordan: Noted. Not sponsored.
Alex: And one more Grok note, since I know a lot of you use the Grok Bot app. Its brains got a serious upgrade last week too. It's now running on Anthropic's Claude Opus five point five, which is one of the smartest models in the world. So Grok Bot is getting the best of both worlds: Claude's reasoning for the thinking, and now Grok's world-class voice tech for the talking.
Jordan: Story two, and this one's about your favorite assistant, Muse.
Alex: Oh, it was a big day for Muse. Two things happened yesterday. First, the well-known tech analyst Gene Munster, from Deepwater Asset Management, came out and said Meta's Muse is a bigger long-term threat to Google Search than ChatGPT ever was.
Jordan: Bigger than ChatGPT? That's bold.
Alex: His argument is actually pretty sharp. He said conversational chatbots like ChatGPT mostly answer questions that Google never really monetized, things like homework help and creative writing. But a personal agent like Muse can carry out multi-step tasks and actually initiate purchases and bookings, and that is exactly what Google built a two hundred fifty billion dollar search business on: routing those transactions.
Jordan: So the threat isn't answering questions. It's doing the shopping.
Alex: Exactly. And the market seems to be taking it seriously, or not seriously enough, depending on how you read it. Alphabet shares are actually up about seven percent since Muse launched, which Munster called surprising, like investors haven't fully priced in the risk yet.
Jordan: Hold on. Munster is an investor. Is he talking his book here, or is this a real read?
Alex: Honest answer: a bit of both, probably. Analysts always have an angle. But the logic holds up on its own. When an AI can research a trip, compare prices, and book the flight, Google's cut of that transaction is at risk, and that's a much bigger deal than losing a few "what is photosynthesis" searches to ChatGPT.
Jordan: Fair enough. What's the second Muse thing?
Alex: Some independent-ish validation of the brains. A company called DualEntry, which builds benchmarks for accounting AI, announced yesterday that Muse Wonder scored eighty-three point two percent on their accounting tasks test, tying for second place out of fifty-seven AI models tested.
Jordan: How big a jump is that?
Alex: Massive. Muse's previous model, Muse Spark one point one, scored sixty-one point four percent. So Wonder jumped more than twenty points. It beat every OpenAI model, the best was G P T five point five Pro at eighty point two, and every Google model. It tied Anthropic's Claude Fable five point one. The only model that scored higher was Grok four point five, at eighty-four point two.
Jordan: That's accounting, though. Spreadsheets and reconciliations. Does that mean it's better at everyday stuff?
Alex: That's the right caveat. This is a business benchmark: financial reporting, bank reconciliation, that kind of thing. It's not a "help me plan my day" test. But it does suggest Wonder's reasoning engine is legitimately frontier-level. And Meta says Wonder rolls out to Muse app subscribers first. So if you're on the paid tier, you should be seeing the good stuff soon.
Alex: Story three lands right on your phone. Google announced yesterday, the fourteenth, that Google Messages is getting an opt-in Enhanced Protection mode.
Jordan: What does that do?
Alex: It's aimed at the slow-burn scams. You know the pattern: a text from an unknown number that looks like a wrong number or a job offer, they build rapport for days, then ask for money or a login. The existing on-device detection misses those because each individual message looks innocent.
Jordan: So now Google reads my texts?
Alex: Only if you opt in, and only in conversations with unknown senders. The feature sends the conversation context to Google's most capable Gemini models in the cloud, and Google says the analysis happens inside something called Private AI Compute, which is encrypted and processed in a secure space that even Google itself can't access.
Jordan: That's Google's own claim, presumably.
Alex: Presumably, yes. They've used the same system for features like Pixel Screenshots. The other half is link checking: on-device Gemini Nano will check unfamiliar links against real-time web signals before you tap them, on the Pixel ten and newer, the Galaxy S twenty-six, the new Galaxy foldables, and select other devices. And they redesigned the warnings, so you get more context and a longer pause before tapping a link from a stranger.
Jordan: When does this land?
Alex: Rolling out in the coming weeks, no exact date. Google says their existing scam detection already flags more than five million suspected scam calls and texts a month. But this is the first time the heavy-duty Gemini models are getting involved in your messages.
Jordan: I have to say, "let Gemini read my texts" is a hard sell, even with the fancy encrypted compute box.
Alex: I get it, and that's exactly why it's opt in. The trade-off is real: the scams this targets are the ones humans fall for most, the patient ones. An encrypted analysis that even Google can't see is about as good a deal as you're going to get. Whether people trust it enough to turn it on, that's the real test.
Jordan: Okay, Amazon time. You said they killed the Fire tablet?
Alex: Pretty much. Yesterday Amazon launched the Alexa tablets, the first hardware built around the AI assistant's name. Three models. The Alexa Tablet eight starts at two hundred thirty dollars, eight point seven inch screen, made for reading and watching. The eleven inch is three hundred thirty dollars, with a two point five K, ninety hertz display and Amazon's own fifth-generation neural chip for on-device AI. And the flagship, the Alexa Tablet twelve Pro, starts at five hundred dollars, twelve inch screen, one hundred twenty hertz, and Amazon says three times the processing power and nine times the graphics of any tablet it has ever made.
Jordan: That's a big jump from the Fire H D ten at a hundred fifty-five.
Alex: That's the point. Amazon is going premium. The headline feature is called Onscreen Intelligence: the assistant can see what's on your screen and do things with it. There's an Ask Alexa button, full smart home control, and personalized recommendations. The middle model, the eleven inch, also gets dual Dolby Atmos speakers and a claimed fifteen hours of video playback, which is genuinely good. It launched yesterday in the U S, Mexico, and Canada, and it comes to the U K and Europe on the nineteenth.
Jordan: So Amazon's answer to the iPad is an AI-first tablet?
Alex: More like Amazon's answer to "nobody talks to their tablet." They want this to be a device you interact with, not just consume on. Whether people will pay five hundred dollars for an Amazon tablet, that's the experiment.
Alex: Story five is for the builders, but it matters if you pay for AI coding tools. Cursor, the AI coding app, changed its plans this week. New customers now start on something called the X Subscription, which bundles Cursor with Grok models and X Premium.
Jordan: And the catch?
Alex: Claude and G P T are no longer included in the plan. You can still use them on the Premium tiers, but you pay API list price from the very first request, out of prepaid credit. Existing subscribers are grandfathered, they don't get moved. But if you sign up this week, you're in the Grok lane.
Jordan: What if I already pay for Cursor? Do I need to do anything?
Alex: Nope. If you're an existing subscriber, nothing changes, your plan stays exactly as it is. This only hits new signups. But it's a signal about where the wind is blowing: Grok is now a first-class citizen in one of the most popular coding tools in the world, and Claude and G P T just became the expensive add-ons.
Jordan: So xAI and Cursor are betting Grok is good enough that you won't miss Claude?
Alex: That's the bet. And speaking of Claude, Anthropic had a good news, bad news day. A security researcher found that Claude Code was sending your API gateway key to Anthropic when it shouldn't have been. The good news: version two point one point two nine nine fixes it. So if you run Claude Code through your own gateway, update today.
Jordan: Quick hits time, and there's a lot today. First, the reviews are in on Apple's new HomeView hub, and Marques Brownlee is not sold. In his launch-day impressions yesterday, he said he couldn't figure out who it's for, and called it, quote, "a skip for most people," saying after two weeks of novelty it just becomes a screen sitting in the corner. Reddit agrees, by the way. Apple's launch ad went viral for the wrong reasons, with users calling it anxiety-inducing and saying the product adds to the chaos instead of solving it.
Alex: Ouch. Okay, next: if your phone got more expensive this year, blame AI. Mrwhosetheboss broke down why the iPhone eighteen Pro, the Galaxy S twenty-six, and the Pixel ten A each went up about a hundred dollars. Memory chips now make up roughly a third of a phone's parts cost, up from a tenth, because phone makers are bidding against AI data centers for the same wafers.
Jordan: So the AI boom is literally raising your phone bill.
Alex: It really is. Next: a rumor, so take it with a grain of salt. Roland Quandt tipped that Nothing is working on two smart glasses: an audio-only pair, and a camera model with ChatGPT built in plus live translation, both with photochromic lenses. Nothing confirmed yet, but ChatGPT in your glasses is exactly where this race is heading.
Jordan: One for the builders: an open-source tool called REA is getting attention after YouTuber Matthew Berman covered it. It lets AI coding agents decompile and understand shipped software without the source code. Berman compared the moment to the early days of Napster, which tells you how wild the implications are.
Alex: And two more. Anker's eufy solar wall light cam, the S four, hit an all-time low of one hundred sixty-nine ninety-nine on Amazon yesterday. Four K video, eight times zoom, full-color night vision, runs on solar, and it works with Apple HomeKit through a hub and Google Home.
Jordan: And a fascinating survey dropped yesterday. Product dot A I asked five hundred fifty shoppers who use AI, and almost ninety percent double-check the AI's recommendations before buying. Seventy-one percent plan to use AI for holiday shopping this season, but only about one in five has ever let an AI actually complete a purchase. The two most used shopping assistants? Gemini and ChatGPT.
Alex: So people want the AI's advice, but they won't hand it the credit card.
Jordan: Not yet.
Alex: All right, time for Try This Today. Today's pick: when Google's Enhanced Protection mode lands in your Messages app in the coming weeks, turn it on. It's opt in, so it won't just appear by itself.
Jordan: Where do I even find it?
Alex: Google hasn't published the exact settings path yet, but the moment it rolls out, we'll flag it on this show. In the meantime, here's what you CAN do today: open Google Messages, go into settings, and make sure basic Scam Detection and spam protection are switched on. On most phones they're on by default, but check anyway. And if you get a text from an unknown number that feels just a little too friendly, slow down before you tap any link. The scammers are patient. Be more patient.
Jordan: Free, takes thirty seconds, and it might save you from losing real money. Good pick.
Alex: Quick recap. Grok Voice three point oh is the new number one voice AI in the world, and it only costs developers eight cents a minute. Meta's Muse is being called a bigger threat to Google than ChatGPT, and Muse Wonder just aced an accounting benchmark. Google Messages is getting Gemini-powered scam protection. Amazon launched premium Alexa tablets. And Cursor is going all in on Grok.
Jordan: Plus the quick hits: M K B H D says skip the HomeView, your phone got more expensive because of AI, and Nothing might be putting ChatGPT in your glasses.
Alex: And that's the show. If you're liking it, make sure to follow or subscribe so you get the latest tech news every day.
Jordan: See you tomorrow morning. I'm Jordan.
Alex: And I'm Alex. Bye, everyone.