Gemini 4 Carbon, Anthropic's IPO Books, and a Cheaper GPT

Grant: Google dropped two new AI models in one week.

Claire: And Anthropic just filed paperwork that opened its books completely. Sixteen point three billion dollars in six months, if you can believe that.

Grant: Meanwhile Google's own search ad growth is slowing down, and the company is actually pointing some of the blame at its own AI search feature.

Claire: There's also a new OpenAI model at half the price, a DoorDash reality check on AI shopping agents, and what's really happening to publisher traffic. Lots to get through.

Grant: I'm Grant.

Claire: And I'm Claire. This is The Week in AI, your Monday morning catch up before the work week starts, covering the week ending Sunday, November first, twenty twenty six.

Grant: So Google had a loaded week. According to Google DeepMind, they announced Gemini four Carbon, their newest frontier model, just two weeks after Gemini four Argon. It's aimed at long running engineering work and cybersecurity defense, and Google says it leads the Artificial Analysis Intelligence Index.

Claire: Two major models in two weeks is fast even for Google. Who's actually allowed to touch Carbon right now?

Grant: Right now it's limited to a subset of cyber defenders through what Google calls the Fairwind Program. More trusted partners get added over time, and broader access for developers and businesses comes once they refine the guardrails. Google's also part of a voluntary government pre release testing process.

Claire: What about Genie four, the world model? TechCrunch had details on that.

Grant: Yeah, Genie four now adds audio, so music, ambient sound, dialogue, and it improves physics. You can change weather, camera angle or objects while the scene is still running, and scenes hold together for up to two minutes now.

Claire: Is that actually usable for making content yet, or is it still a lab demo?

Grant: Honestly, still early. A Google engineer said the end to end product experience is unclear, and pointed to ideation and storyboarding as the best current fits. Ultra subscribers and early API partners get access in the coming weeks, but there's no pricing yet.

Claire: So for marketers watching this, what's the actual move?

Grant: Test, don't plan a rollout yet. And if you're already on Argon, good news, Carbon keeps Argon's introductory pricing, so there's no rate shock when wider access opens.

Claire: Anthropic filed to go public this week under the ticker A N T, and the S one gave us numbers we've genuinely never seen before. Revenue in the first half of twenty twenty six hit sixteen point three billion dollars, up from just one point two billion a year earlier. That's reported by TechCrunch, The Verge, Wired and others.

Grant: That's over a ten times jump. What's the catch?

Claire: The operating loss was six hundred ninety three million dollars, but the net loss was a staggering eighty point seven billion. TechCrunch says that's mostly noncash accounting charges tied to convertible debt the company issued, not straight cash burn.

Grant: Still, that headline number is going to spook people.

Claire: Anthropic even says in the filing it may not reach or maintain profitability. But the customer numbers are the useful part for us. As of the end of September, more than eleven thousand nine hundred customers spend at least a hundred thousand dollars a year, and about two hundred fifty spend ten million dollars or more.

Grant: What about who actually controls the company once shares are trading?

Claire: Seven co founders will hold fifty point one percent of voting power through a Founder L L C, and a Long Term Benefit Trust appoints most of the board. So public shareholders won't have much sway.

Grant: So what should teams running their workflows on Claude take from all this?

Claire: Watch for the amended filing with actual share count and price. And if Claude is part of your stack, the risk factors and spending disclosures are genuinely worth a read as part of vendor planning.

Grant: Staying with Google, they're taking AI Mode ads global now. Search Engine Journal lists ten new markets, including Australia, Canada, India and Singapore, with Brazil, Mexico and the UK coming in the fourth quarter.

Claire: And they're adding ads on follow up questions too. How does that work for advertisers who are already running campaigns?

Grant: If your campaigns already run on Search, Shopping or Performance Max, you're in automatically. There's no separate opt out.

Claire: Can advertisers target AI Mode on its own at least?

Grant: No, you can't target it in isolation. Negative keywords and brand exclusions still apply, but reporting folds AI Mode results into regular Search data, so you can't cleanly isolate performance.

Claire: And this is all happening while Search revenue growth is cooling, right?

Grant: Exactly. Search revenue rose twelve percent in Q three, down from seventeen percent in Q two. Google's chief business officer Philipp Schindler said lower AI Mode monetization was one factor.

Claire: So what do advertisers actually do with that?

Grant: Since you can't isolate the placement, lean on your own conversion data as these ads spread into more markets. And keep an eye on the Q four rollout to Brazil, Mexico and the UK.

Claire: Buried in that same Anthropic filing is something marketers should really notice, their own ad spending. Digiday found it jumped from thirty three million dollars in twenty twenty four to a hundred forty seven million in twenty twenty five, and it kept climbing this year.

Grant: That's a huge jump for a company that insists it won't put ads inside Claude itself.

Claire: Right, Claude stays ad free. This is Anthropic's own marketing and PR spend around product launches. They also say enterprise customers monetize better than consumer users.

Grant: How much of the revenue is usage based versus flat subscriptions?

Claire: Usage fees brought in thirteen point one billion of that sixteen point three billion first half total, and the company expects usage to stay the majority going forward.

Grant: Anything in the risk section worth flagging for us?

Claire: They flag brand perception, getting lumped in with bad press about other AI companies, job displacement coverage, and Wired notes they even cite the government's supply chain risk designation as a threat to sales.

Grant: So what does a louder Anthropic brand mean for agencies and content teams?

Claire: Wired reports, citing the Wall Street Journal, that the actual IPO could land later in November.

Grant: OpenAI dropped GPT six point one Terra this week, and the headline is price. According to The Information, it's half the API cost of the previous Terra model.

Claire: How does that compare to Anthropic's pricing?

Grant: The Decoder ran the numbers. Input tokens are one dollar per million, output is five dollars, against two dollars and ten dollars for Claude Sonnet five point five.

Claire: Is it actually better, or just cheaper?

Grant: Both, apparently. Artificial Analysis found a six point gain on its Intelligence Index over the prior Terra at max effort, with per task cost down to forty cents from a dollar forty.

Claire: What about hallucinations, any movement there?

Grant: On a private benchmark, guessing when the model doesn't know the answer fell to thirty eight percent, down from eighty eight percent for the prior Terra.

Claire: So what should content teams actually do with this?

Grant: Routine agent work, drafting, form filling, document review, gets a lot cheaper to run at volume. Still test it on your own real workflows though, and OpenAI still points to its Sol and Astra models for the hardest coding jobs.

Claire: There's real friction building between publishers and Google's AI search features. The Verge reports that Digital Content Next found nineteen of its members lost between one and twenty five percent of their Google traffic between May and June of last year.

Grant: That's a wide range. Any starker examples in there?

Claire: DMG Media told the UK's competition regulator its click through rates fell by up to eighty nine percent between June and July of twenty twenty five, and they pointed to AI Overviews.

Grant: Google disputes that framing though, right?

Claire: Yes, Google says click volume is relatively stable and that the web is thriving. So the data is genuinely contested.

Grant: If traffic numbers are this disputed, how should marketers even measure success?

Claire: Search Engine Journal covered a webinar where Tom Capper from STAT argued for tracking brand mentions and recommendations as the real outcome, not just domain citations. He says sample your prompts broadly and watch sub brands and competitors too.

Grant: What's a concrete first step for a team that wants to start doing that?

Claire: He suggests post purchase surveys fed into analytics as another dimension, comparing what customers say discovered them against what your channels get credited for. Honestly, start by auditing your current dashboards for where they only count visits and miss mentions entirely.

Grant: Sticking with Google, on the earnings call itself, an analyst straight up asked if AI Mode is eating into ad revenue. According to Search Engine Journal and Latent Space, the answer was basically yes.

Claire: What exactly did Schindler say?

Grant: He ranked currency as the top cause, then retail and media trends, but he also said lower monetization in AI Mode mattered because query growth there is fast and it earns less per query.

Claire: Did he give any actual numbers on how much lower?

Grant: No comparison, no size given. And Google didn't update its click to website numbers this quarter, after sharing that figure just one quarter earlier.

Claire: And that same day, they expanded AI Mode ads anyway?

Grant: Right, ten more countries got ads that very day, plus multiturn ads are coming to the U S. So revenue pressure and expansion are happening at the same time.

Claire: What should marketers watch for next?

Grant: Whether Google starts breaking out AI Mode numbers directly, or restores the click disclosures. In the meantime, check your own analytics for referral shifts from AI surfaces, because Google isn't going to answer that for you.

Claire: Last one is a nice reality check on agentic commerce. DoorDash's C E O Tony Xu said on the earnings call, as reported by TechCrunch and Ars Technica, that third party AI agents have placed only several hundred orders total.

Grant: Several hundred, out of how many overall?

Claire: Ars Technica says DoorDash handled just under one billion orders in the third quarter. So agent volume is basically a rounding error.

Grant: Is the agent connector even widely available yet?

Claire: Not really. Their M C P connector is less than a month old and limited to a handful of early beta partners, mostly for corporate lunch orders.

Grant: Did Xu say why he thinks the number is so small?

Claire: He thinks agents fit bigger, slower purchase decisions better, while a simple dinner order still goes straight to the DoorDash app. His line was, you never want to fight gravity.

Grant: So what's the signal for marketers thinking about agent commerce right now?

Claire: Treat agent specific spending as an early experiment, not a core channel, and watch for whether DoorDash opens the connector wider or builds ad options inside agents.

Grant: Okay, so with all of that, what's the one thing a marketing team should actually do this week?

Claire: Two small moves that take an afternoon. First, pull your Google Ads account and check which campaigns are now eligible for AI Mode placements, since Search, Shopping and Performance Max campaigns are already in automatically with no opt out.

Grant: And second?

Claire: Set up even one brand mention tracking query across a few AI tools, not just a domain citation search. Between Google's own admission about AI Mode and the publisher traffic numbers we talked about, clicks alone just aren't telling the whole story anymore.

Grant: That's the week. A lot of it comes down to the same theme, AI is quietly running more of the work behind marketing, from search placements to agent orders, and the teams keeping up are the ones automating the boring parts and watching the data closely.

Claire: Talk with our team, the links are in the show notes, or call or WhatsApp us at four one nine, five zero zero, four eight eight two.

Grant: And if you just want the key points every day instead of waiting for Monday, subscribe to the AI news wire. Thanks for starting your week with us.

Grant: If you want this kind of automation working for your own brand, Z Media builds fully automated AI content and CRM systems that write, design, film, publish and answer for you. Talk with our team: the links are in the show notes, or call or WhatsApp us at four one nine, five zero zero, four eight eight two. And for the key points every day, subscribe to the AI news wire.

Claire: See you next time on The Z Media Podcast.