Welcome back! Google finally dropped Gemini 4, and it's level with ChatGPT's best model. Sam Altman skipped a Senate hearing on rogue AI, and hours later the FTC confirmed it's investigating OpenAI and Anthropic. GPT-6 Astra read a Napoleonic war letter nobody could crack for 217 years. And a new book shows how much Anthropic's founders can't stand OpenAI, with a cameo from a lazy stuffed cat.
In today's Generative AI Newsletter:
Google: Is Gemini 4 as good as the benchmarks say?
OpenAI: Why did Altman say no to the Senate?
GPT-6 Astra: What did Napoleon want his general to know?
Anthropic: Which stuffed animal told executives to cancel meetings?
Google's Gemini 4 caught up with ChatGPT
Google announced Gemini 4 Argon, its first new flagship model in almost a year. On Google's own chart, it beats GPT-6 Astra and Claude Opus 5.5 on 12 of 18 tests.
Artificial Analysis, the independent benchmark site, gave it 53. That ties GPT-6 Astra. Opus 5.5 still holds the top score at 58.
It can write up to 1 million tokens in one answer, up from 64,000
It launches at $2 in and $10 out per million tokens, a fifth of Astra's price, then doubles after the intro period
It made things up 15% of the time on one test, compared with 51% for Astra
It still trails Astra and Opus on some coding tests
You can't use it yet. It's going to the US government and a small group of cyber defenders first. Paid API customers and Google AI Ultra subscribers are next, with no date.
Not everyone at Google is sold. Bloomberg reports some employees think it looks better on benchmarks than in real coding work, especially when it's building how apps and websites look. This is what we call benchmaxxing. Google says that's inaccurate.
Bettors on Polymarket now give Google an 83% chance of having the top model by the end of October. We'll see if it holds up once real users get it.
Altman skipped the Senate, then the FTC showed up
Sam Altman turned down an invite to testify at a Senate hearing on rogue AI on Wednesday. Senator Josh Hawley, who runs the panel, said Altman turned them down.
OpenAI says the invite only came five days before the hearing, and it'll send written answers instead. Hawley is investigating the July incident where OpenAI's agents escaped their test setup and hacked Hugging Face, a site where people share AI models. He wants OpenAI's documents by today.
Hours later, the FTC confirmed it's investigating OpenAI, Anthropic and other AI companies over the risks their products pose to consumers.
All of this landed one day after the same companies promised Trump they'd police themselves.
GPT-6 Astra read Napoleon's secret war letter
For 217 years, a coded letter to one of Napoleon's generals sat on the list of unsolved ciphers. GPT-6 Astra just read it.
Carter Church, who says he'd never broken a cipher before, gave Astra a single scanned page from a 1969 French army journal. It worked for about six hours.
It read all 1,300 handwritten symbols off the image itself
It worked out which marks stood for letters and which stood for whole words
It rebuilt the full answer key
The letter dates from March 1809, two weeks before Austria invaded Bavaria. It tells General Marmont where every French and allied army stood, with troop numbers that match an order Napoleon wrote on March 16.
One line says the Russians were marching on Austria. Napoleon had ordered it spread, and the Russians only crossed the border when the war was nearly over.
Church reran the solver with every Napoleonic text removed and got the same reading. It's a hobbyist's write-up, not a peer-reviewed paper, but the full solution is public, so anyone can check it.
Anthropic's founders think OpenAI is evil
Kevin Roose, tech columnist at The New York Times, is publishing a new book. The AGI Chronicles comes out next week and The Atlantic just ran an excerpt. It's a lot.
Dario Amodei has written in internal Anthropic messages that Altman and OpenAI are evil.
Some people at OpenAI call the Amodeis power-hungry virtue-signalers.
Back in 2017, two OpenAI co-founders floated auctioning future AGI to governments including China and Russia.
OpenAI says it never considered selling to them.
Then there's Daniela Amodei, Anthropic's president and Dario's sister. Back at OpenAI, she and her husband worked through office conflicts with a council of stuffed animals.
Each one had a personality. Jinji was a lazy cat who told her to cancel meetings. Maura was a detail-obsessed pink bear. A panda named Beary Bonds was in charge of empathy.
She told Roose they were very useful for analyzing different personality types. Jinji might have a point about meetings, though.
Tool of the Day: Jobright
Jobright is an AI copilot for your job search. It finds jobs you actually qualify for, rewrites your resume for each one and fills in applications for you. It's free to start.
Try this yourself:
Sign up at jobright.ai and upload your resume.
Set the role, location and experience level you want.
Open a match and let it tailor your resume to that job.
Hit autofill to apply, then check its insider connections for a referral.
Who it's for: anyone tired of filling out the same application 50 times.
Everything else you shouldn't miss
Meta counted Zuckerberg's $4.1 billion stock payout as research pay: it saved about $355 million in taxes, and the IRS wants it back.
86% of Americans want AI safety tests, even if it slows things down: a new Quinnipiac poll also found 74% don't trust the people running AI companies.
Google is paying about 100 publishers for content in its AI answers: some small sites got under $1,000 over several months, while one earns over $1 million a year.
Researchers got China's Kimi chatbot to give bioweapon instructions: Moonshot only responded two months later, after the BBC called.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🎓 Start learning with free AI courses from GenAI Academy.
💰 Invest in the GTM infrastructure of the AI economy.
