In partnership with
What's Actually Happening
This weekend, two of the richest companies in tech made very public promises about sharing their wealth, and they disagree on how to do it.
Anthropic is putting 5% of its shares into a fund for the public good. Elon Musk saw the announcement and replied that he'll just give stock directly to people, starting with every child in America.
Meanwhile, two models that haven't been announced are showing up anyway. Claude Fable 5.5 appeared in a Microsoft Azure catalog, and OpenAI's GPT-6.1 Terra appeared in Microsoft's code.
There are four stories here, and they share a theme: the biggest moves in AI keep happening before anyone officially announces them.
ARTIFICIAL INTELLIGENCE
🚨 Anthropic Puts 5% Of Itself Into Public Fund
On Saturday, Anthropic announced the Anthropic Public Benefit Fund. Here's how it works:
📦 5% of Anthropic's shares go into the fund, subject to stockholder approval.
📈 It grows with the company. The board can add another 2% for every 50% increase in Anthropic's value above a baseline.
🎯 What it pays for: large donations to nonprofits and "direct distributions to specific populations." Anthropic names AI's effect on jobs as an example of where bolder approaches may be needed.
🚫 What it can't do: none of the equity can be used for lobbying or political giving, in the US or anywhere else.
💼 Where the shares come from: Anthropic says this is equity that would otherwise have gone into growing executive pay packages.
Put that together with the 80% of equity the co-founders pledged at founding and an employee donation match, and Anthropic says about 20% of its shares are now set aside for public good.
How much that is depends on Anthropic's upcoming IPO. Analysts have floated valuations above $2 trillion, and at that price 5% would be around $100 billion. That's our math, and nothing is priced yet.
Still missing: who runs the fund, what the valuation baseline is, and how money actually gets out the door. Anthropic says those details are coming "in the coming months."
🚀 Musk: A Free SpaceX Share for Every Kid in America
Musk replied to Anthropic's announcement on X: "I recommend giving the shares directly to the public. This is what I will be doing."
Then came the promise: "Next year, I will give one of my SpaceX shares to every kid in America. This will be the first of many."
Here's the math:
👶 About 72 million Americans are under 18.
💵 At around $146 a share, that comes to roughly $10.5 billion.
🧮 Musk held about 6.42 billion SpaceX shares as of June 30. Giving away 72 million of them is about 1% of his stake.
SpaceX has been public since June 12 (Nasdaq: SPCX). It listed at $135 a share in the biggest IPO ever.
What we don't know yet: which ages qualify, how a 10-year-old would actually hold a share, and whether this runs through the new government-funded Trump Accounts. SpaceX's COO Gwynne Shotwell has already pledged more than $320 million of her stock to those accounts.
One more date worth knowing: up to 328 million SpaceX shares come out of lockup on October 24, which could put pressure on the price your kid's share is worth.
Your meetings, inside Claude and ChatGPT
Last week's customer call has the details your proposal needs. Right now they're buried in notes nobody can find.
Wispr Flow Notetaker captures your meetings with names, numbers, and product terms spelled right, then brings them into Claude or ChatGPT through a built-in connector. Ask Claude to draft the proposal from your last three calls with a prospect, or pull together what the deal team said about pricing this quarter.
Need something from a call in March? Ask one question across every meeting you've had and get the answer with citations. Get Wispr Flow Notetaker, free to try on Mac and Windows.
🌟 Fable 5.5 Shows Up in Azure
Two weeks ago we put the Fable 5.5 rumors through our three tests, and they failed the "is there a document?" test. Now there's something closer to one.
🖥️ The Azure listing. A screenshot showed Claude Fable 5.5 in Azure's East US 2 model catalog as "Generally available," priced at $10 / $50 per million tokens, with a 1M context window and 128K output. Those numbers match Fable 5.1 exactly, so they may just be placeholders.
🧹 Then it disappeared. LiteLLM added an entry for it, then removed it a day later when Azure queries came back empty.
🍙 A codename. "claude-onigiri-eap" briefly appeared in Artificial Analysis's site with four effort levels (low through xhigh), then vanished within hours. It's the leading candidate for Fable 5.5.
🤝 Partners. Leakers say partners got access around October 15 and that a launch is likely this week.
Our read: this is the strongest Fable 5.5 evidence so far, but a screenshot isn't a launch. Anthropic still hasn't said anything. Its models page lists Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, and that's all.
🌍 GPT-6.1 Terra Shows Up in Microsoft's Code
Today, a leaker reported that Microsoft changed 18 evaluation configs from "GPT-5.6 Terra" to "GPT-6.1 Terra." The change is said to date from around October 16.
Why it's interesting: the GPT-6 lineup has three models so far. Astra is the top tier, Sol is for coding and agents, and Luna is the cheap one. There hasn't been a GPT-6 version of Terra, which in the 5.6 generation was the $2 / $12 workhorse. A 6.1 Terra would fill that gap.
Our read: this one is weaker than the Fable evidence. It comes from a single leaker, nobody has independently verified it, and OpenAI's public docs list only GPT-6.1 Sol. A renamed eval config means something is being tested. It doesn't tell you when it ships.
Top 5 In AI Research 🔬
The stories moving fast beyond today's headlines:
🔐 Claude traced where several NIST elliptic curve seeds came from: Anthropic says its research model linked them to the lrand48 random number generator seeded with Unix timestamps, which answers a decades-old cryptography mystery.
⚛️ Claude Mythos 5 designs quantum circuits: across eight tasks, one error-decoding step came out with about 12x fewer gates.
👩💻 Anthropic publishes its own hiring data: junior technical hires fell from 27% to 9% of the total, even as overall technical hiring grew.
🔍 Anthropic gives METR evaluators insider-level access: six METR staff get the same level of access as internal risk teams, and Anthropic can't block what they publish.
🧠 HybridAgent: agents that think partly in latent space: the preprint reports 10.4% average gains with 35 to 55% fewer generated tokens across 11 benchmarks.
🛠️ Tools That Are Hot Right Now!
🐳 DeepSeek V4.1 Flash: an updated version built for faster agent and coding work, with open weights released.
🔧 Google Harness²: an open-source Google Research project in which an agent rewrites its own prompts, tools and memory for each task.
📜 Claude Code Log: a local, read-only dashboard for searching your Claude Code sessions and seeing estimated token costs.
What's The Recap?
Musk and Anthropic agree that AI's winners should give some of it back. Where they differ is how, and that difference is worth learning to read, because more of these pledges are coming. Use three questions on any pledge. How big is it?Musk's is about $10 billion and Anthropic's could be ten times that. Who decides? Musk decides alone, while Anthropic's fund has a board, with governance still to be written. When does it actually pay out? Both say next year or "the coming months," and neither has paid anything yet. A small promise with clear answers can do more good than a huge one without them. The same habit works on this week's leaks: an Azure screenshot and a renamed config tell you something is coming, but not when, and not what it's worth. In both cases, wait for details before you count on it.
What Do You Rate Today's Newsletter?
Stay building. 🤖