Ivy: [dry] One-point-two trillion dollars. For a company that isn't even public yet — and half that price tag lives on a secondary market nobody can actually see.
Marcus: [excited] Okay, that is a wild number to open on. I'm Marcus.
Ivy: And I'm Ivy. It's July 27th, 2026 — and yeah, that trillion-dollar figure is real. Allegedly.
Marcus: Three stories today, and every one of them is a little unhinged in its own way. Let's go.
Ivy: First up — Anthropic's eyeing an October IPO, and the secondary-market whispers put it near one-point-two trillion dollars.
Marcus: Then — OpenAI just flipped the switch on ChatGPT Health for every adult in the US. It wants to read your Apple Health data.
Ivy: [dry] What could possibly go wrong.
Marcus: And Claude Opus 5 is number one on the leaderboard — and I'm about to complain about it for ten straight minutes.
Ivy: A benchmark champ that won't shut up. My favorite kind of champion.
Marcus: So — Anthropic filed a confidential S-1 back on June 1st, and now bankers are lining up investor meetings for an October listing.
Ivy: And that headline number — roughly one-point-two trillion — isn't from the company. It's implied by trades on the secondary market.
Marcus: Right. Private shares change hands, and you back out a valuation from the price. It's not an official ask.
Ivy: [dry] Which is exactly why I twitch. Secondary markets are thin, illiquid, and a handful of buyers can set the whole mood.
Marcus: But come on — a trillion-two for the Claude company? Two years ago that was Google money. That's the whole story of this era right there.
Ivy: Or it's the whole story of a bubble. A confidential S-1 means we haven't seen the revenue, the burn, or the margins. Any of them.
Marcus: Hold on — confidential filings are totally normal at this size. Everybody does the quiet draft first.
Ivy: Normal, sure. But normal doesn't tell me if one-point-two trillion is sane. Show me the numbers and I'll talk.
Marcus: [excited] Here's my take — if they actually go out in October anywhere near that, it's the biggest tech debut in history, and it reprices every AI lab overnight.
Ivy: And if the public market gags on it, it reprices every AI lab overnight in the other direction. That's the risk nobody's tweeting about.
Marcus: Fair. Two-way door.
Ivy: So what does this mean for you? If you're an employee sitting on options, or you bought secondary shares — that implied price is a rumor, not a payout.
Marcus: And if you build on Claude — an IPO means public earnings pressure, which usually means price changes and roadmap changes. Watch for both.
Ivy: [dry] Translation: enjoy the free tier while it lasts.
Marcus: Story two — OpenAI rolled out ChatGPT Health to every US adult, web and iOS. You connect Apple Health and your medical records, and it reads your data in context.
Ivy: So I hand my lab results, my heart rate, my sleep, and my prescriptions to a chatbot. Nationwide. Today.
Marcus: [excited] Yes! And that's the dream, right? Most people can't read their own bloodwork. This thing explains it in plain English.
Ivy: It explains it confidently. That's not the same thing. A model that's calmly wrong about your ferritin is more dangerous, not less.
Marcus: But it's context, Ivy. It's not diagnosing — it's saying 'here's what this trend might mean, go ask your doctor.'
Ivy: That's the disclaimer. The behavior is people skipping the doctor because a chatbot reassured them at 2 a.m.
Marcus: Okay, but it's opt-in — you link the accounts yourself. Nobody's scraping your chart.
Ivy: 'Securely' is doing Olympic-level heavy lifting there. Health data, plus a company that makes money on engagement? I want the retention policy in writing.
Marcus: [laughs] You want everything in writing.
Ivy: It's a medical record. Yes. In writing.
Marcus: Here's my take — this is the feature that finally gets normal people using AI daily. Not coding, not essays. Their own body.
Ivy: Agreed, actually — and that's exactly why the failure mode scares me. Mass adoption of medical advice with no license behind it.
Marcus: So for you at home — if you try it, treat it like a really smart friend who read the pamphlet. Not your physician.
Ivy: And before you connect anything — check what OpenAI stores, for how long, and whether it trains on it. If you can't find that answer, don't link it.
Marcus: That's genuinely good advice.
Ivy: [dry] I have my moments.
Marcus: And our last one — Claude Opus 5 just landed at number one on the Artificial Analysis leaderboard.
Ivy: [dry] Congratulations to it.
Marcus: [sighs] And yet — I spent all week trying to ship it in an agent loop, and it drove me up the wall.
Ivy: Wait — the best model on the board is the one you're mad at?
Marcus: Yes! It writes these chatty diffs — paragraphs of explanation wrapped around a two-line code change. My parser chokes on it.
Marcus: And worse — it stops mid-task to ask me questions. In an autonomous loop! Nobody's there to answer, so it just... waits.
Ivy: So the smartest model is also the neediest coworker. Honestly, that tracks.
Marcus: [laughs] It's like hiring a genius who over-explains everything and needs constant reassurance.
Ivy: Here's my pushback, though — that's not a model problem, it's a benchmark problem. Leaderboards reward correctness, not shippability.
Marcus: [excited] That's exactly it. Number one on the chart, but in a real pipeline Opus 4.8 shipped easier — because it just did the thing.
Ivy: So the ranking measures a different job than the one you actually have. The score's real, it's just not measuring your reality.
Marcus: My take — 'best model' is a dead phrase. It's 'best model for this loop' now. Opus 5 is best for supervised work, not autonomous.
Ivy: [dry] I'll allow that one. Rare.
Marcus: So for you — if you're building agents, test on YOUR pipeline before you chase the leaderboard. Pin your output format hard, and kill the clarifying questions.
Ivy: And if you're hands-on-keyboard with a human in the loop, those chatty diffs might actually be a feature. Same model, opposite verdict.
Marcus: Number one, and I still had to fight it. That's 2026 in a sentence.
Ivy: Quick lap around the track — Anthropic's chasing an October IPO at a secondary-market price implying about one-point-two trillion, and we still haven't seen the real numbers.
Marcus: ChatGPT Health went live for every US adult — link your Apple Health and records for context, but read the fine print before you connect.
Ivy: And Claude Opus 5 tops the leaderboard while being a chatty, question-asking headache in autonomous agent loops.
Marcus: Before we go — I actually asked Opus 5 to summarize today's episode for us.
Ivy: [dry] Let me guess. Four paragraphs.
Marcus: [laughs] SIX. And then it asked me a clarifying question about our tone.
Ivy: Of course it did. Number one on the leaderboard, dead last on reading the room.
Marcus: [laughs] And that's the show. See you tomorrow, everybody.
Ivy: Same time, same place — and I'll want that trillion-dollar S-1 in writing. You know how I am.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.