Saturday, October 31, 2026
Five Stories • Five Minutes

Happy Friday. Before your coffee cools: a French lab just handed out one of the biggest AI models anyone has released, and Wall Street sent Google a $511 billion message about AI search.

What happened. French AI company Mistral released the weights (the trained numbers that make a model work) for Mistral Large 4, nicknamed “Le Chonk.” The model has about 1 trillion parameters and reads both text and images. Businesses and governments can now download it from Hugging Face, run it on their own servers and adapt it. Mistral is pitching it mainly for cyber defense: it scored 82% on a test that asks a model to reproduce a real software flaw and then patch it.

Why it matters. Most top models sit behind a company's servers, where access can be limited or switched off. Mistral's science chief, Pierre Stock, told TechCrunch that with open weights, access “can't be cut off by definition.” Mistral also says the model uses 47% fewer tokens (the chunks of text AI reads and writes) per task than China's GLM-5.3, a leading open rival, which makes it cheaper to run.

What to do. You won't run this one on a laptop: even the most compressed version needs a computer with 512 GB of memory. If you rely on an IT or security provider, ask whether they use open models like this one and where your data gets processed.

Via TechCrunch

What happened. Alphabet shares fell more than 12% on Thursday, erasing $511 billion in market value, the company's biggest loss on record, Bloomberg reported. Search revenue grew 12% to $63.35 billion, down from 19% growth in the first quarter. Google's business chief, Philipp Schindler, said fast growth in AI Mode queries hurt because those searches earn less money than regular ones.

Why it matters. It's the clearest sign yet that AI answers can eat into the ad business that pays for search. Google's response is more ads inside AI answers: ads in AI Mode now reach 10 more countries, with Brazil, Mexico and the UK next, and US users will start seeing ads on follow-up questions.

What to do. If you buy Google search ads, check this week whether your campaigns appear in AI Mode and compare those results with regular search.

Via The Next Web

What happened. Google's Gemini app can now generate, preview and edit documents, PDFs, slide decks, spreadsheets and code inside a conversation. You can expand a file to full screen and type changes yourself, or highlight a passage and ask Gemini to rework it. PDF editing works on the web and mobile; the other formats, including Google Docs and Sheets, are web only.

Why it matters. It removes the copy-and-paste shuffle between a chatbot and your document, which is where a lot of AI time gets lost. Google says the feature is available to all users in every language the Gemini app supports, so free accounts get it too.

What to do. If you use Gemini, try it on something small: upload rough meeting notes, ask for a one-page PDF summary, then fix a line yourself in the editor.

Via Google Blog

What happened. Xreal launched Aura, wired smart glasses that run Google's Android XR. They connect to a phone-sized computing unit and show apps in front of your eyes. The 256GB version costs $1,279, with a $1,099 launch price through December. Orders are open in the US, Canada, Japan and South Korea, with shipping by mid-November.

Why it matters. Google is leaning on partners like Xreal and Samsung to compete with Meta's glasses. Gemini is built in to answer questions about what's on screen, and you can plug in a laptop over USB-C for a big virtual display or a multi-monitor setup.

What to do. If a portable second screen appeals to you, wait for hands-on reviews before ordering. At 95 grams, Aura is heavier than Meta's 69-gram display glasses, and you carry the computing unit too.

Via TechCrunch

What happened. Anthropic added ready-to-use “artifacts,” interactive tables and overviews, to Claude for Legal, part of Claude Enterprise. They cover routine jobs like checking citations, comparing laws across countries, tracking proposed legislation and checking contracts against a company's own rules. Claude links each statement to its source before a lawyer reviews the result.

Why it matters. Simpler setup has been legal teams' most common request since the product launched in May, Anthropic says. More than 60 integrations, including Bloomberg Law, let teams connect their own databases without uploading files by hand, and built-in settings keep required walls between teams.

What to do. If your firm uses Claude Enterprise, pick one routine task, such as a citation check, and test an artifact on a closed matter before trusting it with live work.

Via The Decoder

Picture a big hospital. When you come in with a sprained ankle, you don't see every doctor on staff; the front desk sends you to the right few. A mixture-of-experts model works the same way. It holds many smaller specialist networks, called experts, and a built-in router picks a handful for each word it handles. Mistral Large 4 has 128 experts and uses four at a time, plus one generalist that always runs. That's how a giant model stays quicker and cheaper to use.

Which would you trust more with your business data: an AI model you run yourself, or one from a big provider, and why? Hit reply and tell us.

Know someone who'd like a five-minute AI briefing? Forward this email. New here? Subscribe at takefiveai.com.