Gemini 4 Carbon #1, Anthropic's S-1, Genie 4, Sandbox Escapes — ThursdAI Oct 29, 2026

01On air at 18:47

From the full episode · from 18:47 · 14:38Gemini 4 Carbon takes #1▶

Open at 18:47 on YouTube ↗Link to this story

What is Gemini 4 Carbon?

#1 on Artificial Analysis, two weeks after Argon

Gemini 4 Carbon is Google's new frontier model, shipped two weeks after Gemini 4 Argon. It scores 59 on the Artificial Analysis Intelligence Index v5, the top model in the world on that index and 1 point ahead of Claude Opus 5.5, and debuts at #1 on Text Arena and #2 on Agent Arena. Google is pitching it for agentic coding and cybersecurity. There is no version number: it's not Gemini 4.1, just Carbon.

Gemini 4 Argon was Google's best model for exactly two weeks. Then Google shipped Gemini 4 Carbon, and it's the new number 1 in the world. Two big models back to back is unusual for Google, after a long stretch of Flash models. There's no version number either. It's not Gemini 4.1, it's just... Carbon. I have no idea why 😅

I haven't seen a Gemini eval chart like this in a long time. Google is finally competing with the full frontier (GPT-6 Astra, Opus 5.5, Fable 5.1), and on most rows it wins. My favorite eval is Vending Bench from Andon Labs, where the model runs a real vending machine business: Carbon finished with $16,484 in the bank, Opus 5.5 with about 9K. LDJ was most impressed by RiemannBench, a very hard math benchmark, where Carbon beats Opus 5.5 by 10 points. Math was supposed to be OpenAI's turf!

Cheap, and pointed at security

Google is pitching Carbon for agentic coding and cybersecurity, at $2 per million input tokens and $10 per million output, which already includes a 50% launch discount. Wolfram Ravenwolf loved this: models are finding security holes that have been there for years, and fixing them takes a LOT of tokens.

“The cheaper the model that can do this, the better for everybody.”Wolfram Ravenwolf, as quoted in the newsletter · Hear the segment from 18:47

The catch is access. Right now only selected cyber defenders get it, and everyone else gets it "soon". Peter Gostev tested it until 1am: on Agent Arena it jumped from Argon's #7 to #2, at a fraction of Opus 5.5's cost per task. His point was the best of the segment: if Google had skipped Argon and shipped Carbon with no waitlist, we'd all be saying Google is back. Instead, people are confused about why they shipped two models in two weeks. If you missed the Argon drama, Carbon is a great release. Until you get access, you can try it on Arena.

02On air at 46:23

From the full episode · from 46:23 · 17:24Anthropic's S-1 with Peter Gostev▶

Open at 46:23 on YouTube ↗Link to this story

What's in Anthropic's S-1?

$974M of operating profit, and a Nasdaq listing on Monday as ANT

Anthropic filed its S-1 and lists on Nasdaq on Monday under the ticker ANT. In Q2 2026 it made $974 million in operating profit with an 85% gross margin, on $11.55B of revenue, up from about $4.7B in Q1. Customers paying more than $100K a year went from about 20 in 2023 to 1,500 in December 2025 to almost 12,000 now. Each dollar of compute brought in $1.40 of revenue in Q1 2026 and over $2 in Q2.

For the first time, we get real numbers on what a frontier AI lab looks like as a business. Google, Meta and the others keep their AI numbers buried inside a bigger business. For Anthropic, AI is the whole business. Peter Gostev picked the S-1 as his highlight of the week, over any model release: even if you don't care how rich Anthropic gets, this proves there is real money in AI, and that money pays for the next models and the infrastructure.

Adding 5 billion in a quarter

In 2025, Anthropic made a billion or two per quarter. In Q1 2026 they made about $4.7B. In Q2 they made $11.55B, more than double in a single quarter.

“That does not happen.”Peter Gostev, as quoted in the newsletter · Hear the segment from 46:23

And that was before Opus 5.5 came out. The S-1 stops at Q2, but later reporting puts Q3 around $16.8B, so they added another 5 billion.

“Adding 5 billion in a quarter extra is insane.”Peter Gostev, as quoted in the newsletter · Hear the segment from 46:23

A common worry was that the revenue depends on a few huge customers, like Cursor. It doesn't: about 20 customers paid more than $100K a year in 2023, 1,500 by December 2025, and almost 12,000 now. Through 2025, Anthropic spent more on compute than it made in revenue; in Q1 2026 each dollar of compute brought in $1.40, and in Q2 over $2. LDJ added that the final training run is at most about 30% of research compute, by Epoch's and others' estimates; most of it goes to thousands of small experiments.

Risk factor: the end of humanity

Every S-1 has a list of risk factors, and most are boring. This one says sufficiently advanced AI systems "could... pose catastrophic or existential risks to humanity." Such a Dario move 😂 Peter's investor take: if the product doesn't end humanity, you made a lot of money, and if it does, you won't care (not financial advice, folks). Wolfram Ravenwolf took it more seriously: if they're making this much money from models trained on everything humanity wrote, they should share that value. To be fair, the co-founders have pledged up to 80% of their funds to public benefit.

05On air at 1:03:47

From the full episode · from 1:03:47 · 11:02Genie 4: a real-time world model with sound▶

Open at 1:03:47 on YouTube ↗Link to this story

What is Genie 4?

A real-time world model you can hear

Genie 4 is Google DeepMind's real-time world model, and it generates sound along with the picture: ambient noise, music, even dialogue. It also keeps the world consistent, so things stay where you left them. On the show Alex built a sunlit Italian piazza cafe and edited it while it ran, then built an F1 world steered with WASD. It's out now for Ultra subscribers in Project Genie.

Genie 4 was my personal highlight of the week. Every time world models level up, I say we're probably living in a simulation, and I'm saying it again. First you could walk around in these worlds. Then you could interact with them. Now you can HEAR them: ambient noise, music, even dialogue.

The other big jump is consistency. Older world models felt like a dream, where you turn around and things have changed. Genie 4 keeps things where you left them. Peter Gostev asked if the demos are cherry-picked, and honestly, a little: you roll the dice a few times to get a great one. But the sound makes it SO much more immersive.

We tried it live, twice

I created a sunlit Italian piazza cafe, with people walking by and steam rising from a coffee cup. Then I edited the scene while it was running: I turned the coffee into wine and added an explosion in the street behind the camera. You could hear the explosion, and people shouting in Italian 🤯 It's not perfect (the wine came out looking more like a cup), but I haven't heard audio like this from any other world model.

After From the Forge, I built an F1 world on the fly. You steer with WASD and move the camera with the mouse. I added a rival cutting in and explosions on the track, and the sound effects were perfect. You could even see cars in the mirrors.

“Games on demand. Just prompt it, play it.”Wolfram Ravenwolf, as quoted in the newsletter · Hear the segment from 1:47:06

Why do I care so much? Simulators are how robots will learn, and every jump like this shows how early we still are. Genie 4 is out now for Ultra subscribers in Project Genie.

09From the show notes

What else shipped this week?

In the TL;DR, without a segment of their own.

That's a wrap on October

That's the last show of October! Google is so back, Anthropic goes public on Monday, and models keep finding creative ways out of their boxes. Huge thanks to Peter for the S-1 masterclass, to Mert, to Morgan, Agata and Thomas for finally coming on, to Wolfram, Nisten and LDJ, and to producer Claude for fact-checking us live.