In partnership with
Welcome to OHMO AI. Google is taking the better Gemini models away from free accounts, and the change lands in days. Mistral says Europe now has a frontier model it plans to give away.
The Pentagon says it has stopped using Claude, months after it said it would. And DeepSeek is raising more money than it set out to.
In today's newsletter:
Google limits free Gemini accounts to Flash-Lite from October 9
Mistral previews Mistral Large 4 and promises the weights this month
The Pentagon says it has stopped using Anthropic's Claude
DeepSeek is set to raise more than it set out to ahead of a listing
LATEST IN AI
Free Gemini accounts drop to Flash-Lite from October 9
Image source: Google
OHMO AI: Google changed what a free Gemini account gets. From October 9, anyone on a free plan is limited to Flash-Lite, the model Google describes as its most cost-effective, with Flash and Pro moving behind paid tiers. The company set out the changes on a support page.
The details:
Free users can currently pick from Flash-Lite, Flash and Pro. After the change, the standard Flash model needs a Google AI Plus subscription at $4.99 a month.
Google AI Plus subscribers lose something as well: Gemini 3.1 Pro is coming off that tier too.
Pro and the "Deep Think" reasoning option will be restricted to Google AI Pro or Ultra subscribers, at $19.99 or $99.99 a month.
Google says Pro excels at complex maths and coding prompts and has a "deeper understanding across text, files, images and videos" than its other models.
Why it matters: The app stays. A free account still gets an answer once the change lands, from the model Google rates on cost rather than ability, and what the cheapest subscription now buys is the Flash model free users had last week. For a free user, $4.99 a month is the price of standing still.
PRESENTED BY DELL
Built for the unstoppable.
The Dell Pro powered by Intel® Core™ Ultra with Intel vPro® adjusts power usage based on how you work, so your battery life never slows you down.
MISTRAL AI
Mistral previews a trillion-parameter model and promises the weights this month
Image source: Mistral
OHMO AI: Mistral released a public preview of Mistral Large 4, a model it says will be the most capable open-weight model built outside China once its weights are out. Open-weight means the model's underlying files are published for anyone to download, use and customise, and Mistral says that happens by the end of the month. For now the preview is an API on Mistral Studio, aimed at developers.
The details:
It is a 1 trillion-parameter model with 49 billion active parameters, and Mistral calls it natively multimodal, meaning it was built from the start to handle more than one kind of input.
Mistral says it trained the model from scratch over two months in its own European data centres on Nvidia Grace Blackwell GPUs; its own announcement puts the count at 3,800, while the figure it gave at launch was 4,000.
Chief scientist and co-founder Guillaume Lample called it "a new generation of models" at the launch press conference, where it got the nickname "Le Chonk".
The model will be offered in several regions, including a European deployment Mistral operates itself from start to finish, with no other digital service provider involved and under European law.
Why it matters: Mistral is selling a place as much as a model: a European region, run from its own data centres, where the data sits under EU law. That choice sits with whoever buys the deployment, not with the person typing into it. The trained-from-scratch claim points the same way: the US government has accused Chinese labs of closing the gap by training on the outputs of OpenAI's and Anthropic's models, and Mistral is saying its answer to China carries no such debt.
ANTHROPIC
The Pentagon says it has stopped using Anthropic's Claude
Image source: OHMO AI
OHMO AI: The US Department of Defence says it is no longer using Anthropic's AI tools, an official confirmed this week, months after the department designated the company a supply chain risk on national security grounds. The confirmation follows interviews in which several people said Claude was still being used widely inside the Pentagon.
The details:
Secretary of Defence Pete Hegseth labelled Anthropic a supply chain risk in February and announced then that the Pentagon would stop using it by late August.
Pentagon employees kept using the Claude tools, including Anthropic's Mythos models, "throughout the whole period of the controversy", one of the people said.
The statement came on Monday, in response to a direct question about the department's use of Anthropic.
Why it matters: The order came from the top in February and carried an August deadline; the people using Claude carried on past it, Mythos models included. The deadline came and went. The statement came this week, after a reporter asked.
PRESENTED BY SPRINTO
Your SOC 2, handled end to end with AI agents.
Enterprise buyers will not put your product near their customer data without a SOC 2 report. Sprinto gets you audit ready in 14 days, across three working sessions. AI agents do the collecting, you approve. Your auditor signs off.
DEEPSEEK
DeepSeek's funding round is blowing past its own target
Image source: OHMO AI
OHMO AI: DeepSeek is close to raising at least 80 billion yuan ($12 billion) in its latest funding round, a figure that could rise, with Tencent and battery maker CATL reported among the largest backers. The Chinese lab had set out to raise about 50 billion yuan, but interest outstripped expectations after the release of its latest model, people familiar with the matter said, and with at least 80 billion yuan now in sight it is eyeing an initial public offering in early 2027.
The details:
The round could grow to as much as 100 billion yuan ($14.9 billion), double the earlier target, with state-backed funds and corporate investors competing for stakes.
DeepSeek is seeking a valuation of about 500 billion yuan, or $75 billion; earlier reports had put it at about $74 billion.
The details are not public, and the figures come from people who asked not to be named.
DeepSeek's move to open up to outside funding was seen as a sign of how much capital AI now demands and of pressure to retain talent as competition intensifies.
Why it matters: Opening the door to outside money was read by the people describing the round as a sign of ballooning capital demand and mounting pressure to retain talent. On that reading, a lab that had funded itself is paying up to keep its own people as much as to build, and the listing it's said to be eyeing for early 2027 is how those backers would get their money back.
AI GUIDES
Learn a topic properly by making ChatGPT or Claude teach it instead of answering
Ask a question and you get an answer you forget by tomorrow. Switch on the learning mode first and the AI asks what you already know, explains in short sections at your level, and quizzes you until it sticks. In ChatGPT open the tools menu and pick Study and learn; in Claude open the style dropdown and pick Learning. Nothing to install, and you can turn it off mid-conversation when you just want the answer.
Copy this, swap in the bracketed parts, and paste it in.
I want to learn [topic, e.g. how compound interest works]. I am a [complete beginner / know the basics / fairly confident] and I need this for [why, e.g. an exam next week / a decision at work / curiosity]. Do not just give me the answer. Ask me how I would approach it first, then teach me in short sections, check my understanding with a quick quiz after each one, and tell me what I got wrong and why.The mode makes the AI guide rather than tell, and the prompt gives it your level and goal so the questions, sections, and quizzes fit you. You remember what you worked out yourself far better than what you were handed.
COMMUNITY
Featured Reader
Every newsletter, we feature how a reader is using AI to work smarter, save time, or make life easier.
Today's reader is Gareth P. from Wrexham, United Kingdom:
“Match fishing the Dee and the Llangollen canal since I was nine. Peg draw is luck (you pull your swim number out of a bag), feeding isn't, and I wanted to know which was costing me.
Catch log in Excel, one row per match: venue, peg, water temp off the club thermometer, bait, feeding rate, weight.
Copilot in Excel, question is which columns move the weight most, leave venue out. Forgot to say the log goes back to 2023 so it had enough to go on.
Answer was feeding rate on cold days, not the peg. Groundbait is the stuff you throw in to hold the fish.
ChatGPT with the club byelaws PDF uploaded for what's allowed on each water, page number demanded, it got hooklengths wrong once so I read the page.
AGM's in November and the match sec wants my sheet, dark by seven now so plenty of evenings.”
Want to be in the next newsletter? Tell us how you use AI here.
AI Meme
Go Deeper
AI Guides: Practical AI guides, prompts and tools.
Instagram: The AI page they pray you’ll never find.
Facebook: For the ones who never left Facebook.
