THU · NOV 19, 2026Superintelligence.LONGEVITY
THU · NOV 19, 2026LONGEVITY
 
TODAY’S LEAD
Google is back: Gemini 4 Neon reaches the frontier
Google's new model beats Claude Opus 5.5 on an independent ranking and tops the coding-agent list
READ THE LEAD STORY →
Google
By Kim IsenbergForwarded? Subscribe free →

In partnership with

The Signal · From the editor · Kim Isenberg
Anthropic starts trading on Nasdaq today, and Google just made its job harder.
$83
Anthropic's IPO price per share, set late on Wednesday
$1.5 trillion
Anthropic's valuation as it starts trading on Nasdaq today
61
Gemini 4 Neon on Artificial Analysis's index, ahead of Opus 5.5 at 59
Anthropic priced its shares at $83 late on Wednesday. That values the company at roughly $1.5 trillion and makes it the biggest IPO ever. On the same day, Google released Gemini 4 Neon. On an independent ranking by Artificial Analysis, Neon scores 61 points, ahead of Anthropic's Claude Opus 5.5 at 59, and an average task in that test costs about a third less. Anthropic's best model, Fable 5.2, still leads with 65. But the new shareholders are buying into a company whose customers now have a strong, cheaper option. According to The Information, Anthropic is already offering discounts of more than 30% to keep them.
All the best,Kim Isenberg signatureKim Isenberg
01  ·  IN THE NEWS
BUSINESS  ·  WIRED
Nvidia's interim CEO promises raises after Huang's exit
Wired's report on Colette Kress's first all-hands meeting. Image: Source: WIRED, "Inside Nvidia's First Post-Jensen All-Hands: Pay Raises and More Equity", 11/18/2026; highlight added
→
IN SHORT
Nvidia is paying more to stop its engineers from leaving after Huang's exit, and it comes across as buying their loyalty.
Nvidia's interim CEO Colette Kress told staff at her first all-hands meeting on Wednesday that pay raises and extra stock are coming in December, months ahead of schedule. Pay was the most common of 700 questions employees sent in. Many engineers stayed at Nvidia because of Jensen Huang, who cofounded the company and ran it for 33 years. Coming a week after his exit, the early raises look a lot like an attempt to buy their loyalty. Thousands of employees have reportedly lobbied for Huang's return, while the board looks for a new CEO.
READ AT WIRED  →
MACCUS
The world now needs more cooling than heating, IEA says
Electricity used to cool buildings rises from about 3,000 to 4,700 terawatt-hours a year by 2035 in the IEA's main scenario. Image: MACCUS chart, data: IEA World Energy Outlook 2026, CC BY 4.0
→Air conditioning and AI data centers are both driving up demand on the world's power grids, and the grids have to grow fast to keep up.
For the first time in history, the world's demand for cooling buildings is overtaking the demand for heating them, the International Energy Agency (IEA) said today in its World Energy Outlook. Rising incomes and hotter summers drive the shift, from India to Europe. The same report expects data centers, pushed by the AI boom, to roughly double their electricity use to about 1,000 terawatt-hours a year by 2030. In several regions, new data centers waiting for a grid connection already exceed today's peak demand.
READ AT MACCUS  →
THE INFORMATION
Anthropic offers big discounts as it goes public
The Information cites managers at two customers on Anthropic's discounts. Image: Source: The Information, "Anthropic Is Pushing New Products, and Big Discounts, to Justify Its Huge Valuation", 11/19/2026; highlight added
→Anthropic goes public today as one of the world's most valuable companies, and it is cutting prices to keep its customers from switching.
Anthropic is offering some business customers discounts of more than 30%, The Information reports. In return, they must give Anthropic a fixed share of their AI budget, and Anthropic has started asking for invoices as proof. The company needs to grow beyond coding, with tools like Claude Cowork for office work, to justify the roughly $1.5 trillion valuation it got when it priced its IPO at $83 a share late on Wednesday.
READ AT THE INFORMATION  →
ADVERTISEMENT

Discover tomorrow’s trends today.

Keeping up with industry trends is a process.

TheFutureParty makes it easier with its daily newsletter. 

They deliver the latest stories and insights to help you understand the future of tech, business, and culture — all in a quick, witty package.

02  ·  VIDEO OF THE DAY
Watch: Software engineering after code becomes cheap with Satya Nadella
THE PRAGMATIC ENGINEER
Software engineering after code becomes cheap with Satya Nadella
Microsoft CEO Satya Nadella sat down for an hour with Gergely Orosz of The Pragmatic Engineer. His main point: "In a world where code generation has become cheap, what has become hard is verification." He also answers whether Microsoft's exit offer, open to about 60,000 employees, is the start of AI layoffs: "I wish I could just sit here and say it'll all be the same. I cannot."
WATCH ON YOUTUBE  →
03  ·  QUOTE OF THE DAY
I'm a great admirer of what he's done, and at this point, I kind of want to give him space to figure this out.
Satya Nadella
CEO OF MICROSOFT
On Jensen Huang at the New York Times DealBook Summit
Microsoft
Nadella added that Nvidia's interim CEO, Colette Kress, worked at Microsoft for years: "She's a tremendous person."
SOURCE: CNBC: JENSEN HUANG EXIT FROM NVIDIA IS 'ALL SORT OF PRETTY TRAGIC,' MICROSOFT'S NADELLA SAYS →
04  ·  RUMOURS, LEAKS & DUSTUPS
Moonshot AI, the Chinese maker of the Kimi models, has reportedly scrapped a $150 million deal to rent Nvidia's top Blackwell chips in a data center in Indonesia.
According to The Information, the deal fell apart after a new US rule this week closed the loophole that let Chinese labs use banned chips abroad.
Moonshot AI
THE INFORMATION, EXCLUSIVE
The Information
THE LEAD STORY2 min read
GOOGLEMODEL RELEASE
Google is back: Gemini 4 Neon reaches the frontier
Google is back!
61
points on Artificial Analysis's index, two more than Claude Opus 5.5
$6.44
per average task in that test, against $26.54 with Fable 5.2
17
interactive demos Neon built in my early-access test
Google
THE TAKEAWAY
→Google released Gemini 4 Neon on Wednesday, in the Gemini API and for paying Pro and Ultra subscribers.
→On Artificial Analysis's independent ranking, Neon scores 61, behind Fable 5.2 (65) and GPT-6.2 Astra (62) but ahead of Claude Opus 5.5 (59).
→It is cheap to run: an average task in that test costs about $6.44, against $26.54 with Fable 5.2.
→In my early-access test, it built 17 interactive demos and fixed three broken programs until every test passed.
Google is back! On Wednesday, Google released Gemini 4 Neon, its highest-scoring model yet. On Artificial Analysis's independent intelligence ranking, Neon scores 61 points, two more than Anthropic's Claude Opus 5.5. Only Anthropic's Fable 5.2 (65) and OpenAI's GPT-6.2 Astra (62) score higher. Neon is also efficient. An average task in that test costs about $6.44 with Neon, about a quarter of what it costs with Fable 5.2.
Google's own chart plots the Artificial Analysis score against the cost per task. At every thinking level, Neon (blue) scores higher than Claude Opus 5.5 (dark gray) for less money. Image: Google DeepMind
Google built Neon for long jobs that an AI agent works through on its own, in coding, office work and science. In Google's coding app Antigravity, Neon can also act as a manager that hands smaller jobs to the faster Hydrogen model.
I had early access and tested it for a full day. Neon built 17 interactive demos for me, from a 3D shooter to a jet flight through a canyon with a timer. What impressed me most was its sense of design. For a voxel arena, I only asked for destruction and a reset. Neon added three structures and three tools that each behave differently.
Terra Breaker, the voxel arena Neon built for me: a castle mid-blast, with three demolition tools and three structures to choose from. Image: X / @kimmonismus, 11/18/2026
I also gave it three broken programs. After its fixes, every test passed, and Neon flagged the two small gaps that remained on its own.
My coding test from the early-access thread: three broken programs, all tests passing afterwards. Image: X / @kimmonismus, 11/18/2026
It was not flawless. During my test, the servers returned 57 "no capacity" errors, and four runs stopped until I restarted them. That may have been the load before the launch. On the overall ranking, Fable 5.2 also stays clearly ahead. So Google is back at the frontier: slightly ahead of Opus, cheaper and with real taste, but not yet at Fable's level. But I guess they are still cooking.

Again: Google is back!
→
WHY IT MATTERS
Google has done what OpenAI tried to do with GPT-6.2 earlier this week: release a model that can really compete with Anthropic, right before Anthropic goes public. Anthropic's shares start trading on Nasdaq today at a valuation of about $1.5 trillion. Its business customers now have a strong option at Opus level that costs less.
READ AT GOOGLE  →
ADVERTISEMENT

The voice-to-text app that built its own speech model

Most dictation apps are a wrapper around someone else's speech model. Wispr Flow trained its own.

It's called Canto, and it was built for the way people actually talk to AI: thinking out loud, changing direction mid-sentence, half-whispering in an open office.

Wispr tested it on 10 hours of real dictations from more than 2,300 speakers. Canto had the lowest word error rate of every model tested, including models from Google, OpenAI, AssemblyAI and Deepgram. That means your prompts to ChatGPT, Claude and Cursor come through the way you said them.

Talk naturally in any app and Wispr Flow types clean, formatted text, 4x faster than typing.

05  ·  GRAPH OF THE DAY
FABLE 5.2
58
→
GEMINI 4 NEON
59
Artificial Analysis Coding Agent Index
→  The chartArtificial Analysis's Coding Agent Index tests complete coding tools, each running its maker's own model. The tasks are real programming work: fixing software, working in a terminal and answering technical questions about code. Google's Antigravity with Gemini 4 Neon scores 59, one point ahead of Anthropic's Claude Code with Fable 5.2 (58). OpenAI's Codex with GPT-6.2 Astra follows with 53.
→  The lessonIn coding, the order from the Featured Story flips, and Neon leads Fable 5.2. It also gets there faster and cheaper. An average task takes Neon about 31 minutes and $6.86. Fable 5.2 needs about 53 minutes and $15.47. For coding agents, Google now delivers top results at less than half of Anthropic's cost per task.
→  The caveatThe lead is small: Neon's exact score is 58.6, and Fable's 58 is rounded too. The index also tests each model inside its maker's own tool, so it compares complete products rather than bare models.
SOURCE: ARTIFICIAL ANALYSIS VIA X  →
Daily Feature: Thursday, Longevity
AI can read your heart's real age from an ultrasound
→  Bottom lineA Mayo Clinic AI estimated heart age from routine ultrasound clips, and older-looking hearts were linked to more deaths over 20 years.
→  Why it mattersA standard heart scan could flag people whose hearts are aging faster than they are, long before symptoms appear.
→  What it meansAging becomes measurable organ by organ, so researchers can test whether exercise or drugs make a heart look younger.
568
adults aged 50 or older, followed for 20 years
61%
higher risk for every five extra years of heart age
Your heart has its own age, and it does not always match your birthday. Researchers at the Mayo Clinic built an AI that estimates that age from an echocardiogram, the standard ultrasound scan of the heart. On November 13, they presented what that number is worth at the European Society of Cardiology's Digital & AI Summit. A heart that looks older than its owner points to a higher risk of dying from heart disease or dementia, even 20 years later.
An echocardiography machine of the kind used for the scans: the AI reads the same video clips a cardiologist looks at. Image: Dr D. Malleswara Rao / Wikimedia Commons, CC BY-SA 4.0
The team used scans from 568 adults aged 50 or older, recorded between 2002 and 2004, and let the AI estimate the age of each heart. Then they looked at who died of heart disease or dementia over the next 20 years. Every five extra years of heart age went along with a 61% higher risk, after accounting for age and sex. When the researchers also accounted for weight and how strongly the heart pumps, the link was weaker but still clear, at 46%.
The key result in the ESC press release. Image: Source: European Society of Cardiology, "Having older AI-estimated heart age than actual age is linked with cardiovascular/dementia-related death", 11/13/2026; highlight added
Why dementia? Doctors increasingly see heart disease as a driver of decline in old age, including dementia. The team counted both causes of death together for practical reasons and wants to study them separately next. Two of the researchers have licensed their algorithms to UltraSight, a company that makes AI ultrasound software, and could profit from its use.
For longevity research: The AI's estimate was checked against 20 years of real deaths. The next step is larger studies that follow people forward in time. Then comes the question everyone in the field cares about: can exercise or medication make a heart look younger on the scan?
READ AT ESCARDIO  →
WANT MORE? 
The full Saturday DeepDive every week, plus discounts on AI software and member giveaways.
JOIN SUPERINTEL+  →
$10 a month or $100 a year. Students free with a .edu address.

Reply

Avatar

or to participate