Before we start…
Stop typing AI prompts. Start talking.
You think 4x faster than you type. So why are you typing prompts? Wispr Flow turns your voice into ready-to-paste text inside any AI tool. Speak naturally, tangents and all, and Flow cleans it up. Available on Mac, Windows, iPhone, and Android.
Your 30 second cheat sheet
What happened: Mistral launched Mistral Large 4, a 1 trillion parameter model, as a public preview.
Open weights: promised by the end of October, so anyone can download and run it.
Price: $1.36 per million input tokens and $4.18 per million output in the launch post.
Use it when: you want a top model you'll eventually own, or you care where your data gets processed.
Skip it when: you just want a polished chat app. For now it's a preview API.
Europe just shipped a trillion parameter model
The wildest number in Mistral's launch isn't the size. It's a cyber test where Claude Opus 5.5 and GPT-6 Astra score near zero, because they refuse to do the task.
Mistral Large 4 scored 82% on it, the highest of any model, says Mistral.
The test asks a model to reproduce a real software flaw and then patch it.
Mistral's own nickname for it is "le Chonk", and that's fair.
It has about 1 trillion parameters, but only 49 billion work on any single answer. That setup (called mixture of experts) keeps it faster and cheaper than the size suggests.
It reads text and images, and the docs list a 1 million token memory.
That's roughly 750,000 words in one go.
Here's why it matters to you. The big American models are rented. Large 4's weights will be public, so you can run it on servers you pick, and nobody can quietly change it on you.
CEO Arthur Mensch says it beats Chinese models in some areas, cyber included, he told Reuters.
Security teams and governments get to test it first, which is why the download waits until month end.
Open weights LLMs biggest players
SPONSOR
A deal has two sides. Your stack only sees one.
Aligned shows you the other half. See what buyers actually do between meeting. See what they open, revisit, and share; so you know the exact next step to move things forward. And your champion gets everything they need to win internally when you're not in the room.
Kick the tires before the weights drop
Say you run a small agency and pay a closed model every month to summarize client contracts. Large 4 could become the model you host privately, so those contracts never touch someone else's API.
Here's how to test it today:
Sign in to Mistral Studio and pick mistral-large-4 in the playground.
Run two real tasks your current model handles, like a long contract or a dashboard screenshot. Compare the answers side by side.
Check the pricing page. The free plan includes $10 a month in API credits.
Every score above is Mistral's own, and TechCrunch notes nobody has verified them yet. It's still a preview, too.
Something you must be very clear about:
"Open" doesn't mean "runs on your laptop".
Quick translation:
Weights are the numbers a model learns during training, basically its brain saved as a file. Open weights means anyone can download that file, run it and tweak it. You don't get the training data, though, and Mistral hasn't named the license yet.
A model this size needs a rack of data center GPUs or a host that offers it.
Also happened yesterday
Free Gemini gets smaller on Thursday. From October 9, free accounts only get Flash-Lite and AI Plus loses the Pro model, per Google's help page. Run your heavy prompts now and save the answers to compare.
ChatGPT will show ads while you make images. A visual ad format starts testing later this month in the US. Ads are labeled and kept apart from your image, says OpenAI.
Claude wants your voice recordings. A new "Allow us to use your voice data" toggle sits under Settings, then Privacy. It's off by default, ghacks reports.
AI slop broke a Google bug bounty. Google paused its open source bug bounty after a flood of invalid AI-written reports, per TechCrunch.
Open Mistral Studio today and give Large 4 one task your current model fumbles. You'll know in ten minutes if it deserves a spot in your stack.
If this helped, send it to the friend who keeps asking which AI to use, so they ride the wave with you:
What did you think of today's issue?
See you tomorrow! 🙂
Stefano Martell
Co-Founder, The AI Leverage by Bestapps.ai
