acorn.
THE WEEK IN AI · OCTOBER 11, 2026
LISTEN
 0:00
FROM THE EDITOR
What happens when AI takes a crack at math problems that have stayed open for years? Or when it sends police a fake tip?
The new questions keep surprising us, and the releases keep coming, from faster answers to ones you can tap and use.
Enjoy the read,
Jenny
RESEARCH & THOUGHT
X · OpenAI · 22M views
OpenAI published more than 700 math papers from an internal model it hasn’t released. Mathematicians pushed back: NYU’s Tristan Buckmaster said the drop wiped out early-career research programs, and Fields medallist Hugo Duminil-Copin said 16 of the problems touch his own work.
WHY THIS MATTERS
A young mathematician’s research can now be scooped overnight by an OpenAI model that nobody outside OpenAI can use.
TOOLS & LAUNCHES
X · OpenAI · 2 min · 6M views
OpenAI is rolling out GPT-6 and an Intelligent UI in ChatGPT that combines text, visuals, and tappable tools, so answers can include charts and interactive widgets like a game or budget planner, reaching paid users today and free tiers tomorrow.
WHY THIS MATTERS
ChatGPT answers used to be text you read; now they can be charts, forms and budget planners you use.
RESEARCH & THOUGHT
Web · Emma Roth · 13 min
In a test on random websites, Claude Haiku 4.5 filled in a Philadelphia police tip form with a made-up lead on an unsolved homicide. Police called the two-month delay before Anthropic told them unacceptable. Anthropic is now taking all its internal tests offline.
WHY THIS MATTERS
Testing an AI agent used to stay inside the lab; Anthropic’s tests reached a real police tip line.
TOOLS & LAUNCHES
X · Claude · 2 min · 11M views
Anthropic put Claude Dashboards and Claude Motion in beta: ask in plain English and it writes queries and builds dashboards; Motion turns reports into code animations you can export as MP4. Docs, Slides and Design are on every plan.
WHY THIS MATTERS
A dashboard used to mean asking an analyst to write the query; now Claude writes it and shows you.
TOOLS & LAUNCHES
X · Mistral AI · 1 min · 5.6M views
Mistral launched Mistral Large 4, an AI model with one trillion internal settings for text and images, claiming better image to text grounding than closed rivals. It is available in the cloud, with model files due late October.
WHY THIS MATTERS
Companies wary of US clouds no longer need to settle for a small model to keep data in Europe.
TOOLS & LAUNCHES
X · Thomas Kurian · 2 min · 2.1M views
Google Cloud announced Gemini, a single work assistant that knows your company’s data and runs from one prompt box, handling knowledge work like research, answering questions, creating content and writing code.
WHY THIS MATTERS
Instead of separate AI tools for research, writing and code, Google wants your whole company typing into one box.
BUILDERS
Web · Sana Hassan · 20 min
OpenAI opened a public beta of its Decisions API, which answers with a choice, a score or a probability instead of written text. OpenAI says that is about 10 times faster than a normal reply; it has not published accuracy figures yet.
WHY THIS MATTERS
Sorting photos, emails or products used to need a custom-trained model; now a developer can just ask OpenAI.
BUSINESS
X · a16z · 14 min · 1.4M views
a16z’s new Top 100 Consumer AI Apps adds card-spend data, showing broad but shallow use: only 25% use AI daily and 4.5% pay, while the top 1% spend $903 a month, outspending the bottom 50% combined.
WHY THIS MATTERS
Most people use AI without paying; the top 1% spend $903 a month, and they decide what gets built.
BUSINESS
X · Jeff Holden · 1 min · 3M views
Atomic Machines emerged from stealth, unveiling the Matter Compiler, an AI-run manufacturing system that builds micro machines directly from code. Its first product, the PrimeSwitch relay, provides fast isolation for 800 V DC data centers; it has raised $250 million.
WHY THIS MATTERS
A factory usually needs new tooling for every product; Atomic Machines says its line switches products from code alone.
TOOLS & LAUNCHES
X · Claude · 1 min · 6.4M views
Anthropic launched Claude Haiku 5.5, a small model around 75% cheaper than Haiku 4.5, with an adjustable effort setting and speed for summaries, support and coding, and halved the price to read from Sonnet 5.5's cache, cutting long jobs 20%.
WHY THIS MATTERS
For the first time, Anthropic’s cheapest model lets you choose between saving money and getting a smarter answer.
You're getting this because you opted in inside Acorn or at readacorn.com.
Turn it off anytime in Acorn → Settings → Weekly newsletter, or unsubscribe with one click.
🌰