// this week in AI
What came out in AI this week (and what you actually need)

Every week dozens of AI releases come out. 95% of them are no use to you. Here you'll find the 5% that matters, explained without tech jargon — with a single question in mind: what changes for anyone running a business?
1 · Kimi K3: the world's largest open model
Moonshot AI has released Kimi K3: 2.8 trillion parameters, it understands text, images and video, and it has a one-million-token “memory” — in practice, you can feed it entire archives. Starting July 27, the weights will be fully open.
2 · Google launches Gemini 3.6 Flash
On July 21 Google updated its “Flash” family: models designed to be fast and cheap, not to win benchmarks.
3 · Karpathy moves to Anthropic
Andrej Karpathy — one of the most respected researchers in the world, formerly of OpenAI and Tesla — has joined Anthropic (the company behind Claude) to work on frontier models.
4 · OpenAI heads for the stock market
OpenAI is preparing a confidential IPO filing, at a private valuation of around $730 billion. A listing could come as early as September.
5 · French antitrust turns the spotlight on AI agents
France's competition authority has published a 3,700-page opinion: three companies (OpenAI, Google, Anthropic) control over 84% of the AI agent market.
In short
Open models getting more powerful, fast models getting cheaper, and a market that's consolidating. The lesson of the week: you don't have to chase every new release — you have to choose the right mix for your business. And the right mix changes more slowly than the headlines.
Frequently asked questions
What is Kimi K3, and why does it matter if you run a business?
It's the world's largest open model, released by Moonshot AI: 2.8 trillion parameters, it understands text, images and video, and it has a one-million-token memory. “Open” means the cost of high-end AI keeps falling and that, if you handle sensitive customer data, you can run it on your own servers without sending anything out.
Do I need the most powerful AI model for emails and summaries?
No. For 80% of everyday tasks, like emails, summaries and replies to customers, you don't need the smartest model but the most cost-effective one. That's the point of Gemini 3.6 Flash, updated by Google on July 21: models built to be fast and cheap. The right rule is the right model for each job, not the most expensive one everywhere.
When is OpenAI going public, and what changes for people who use AI?
OpenAI is preparing a confidential IPO filing, with a private valuation of around $730 billion and a listing possible as early as September. For you, it means AI is becoming infrastructure, like electricity or the cloud: prices and products that look more and more like a utility, reliable, but with price lists you need to understand before you sign.
A half-hour chat, no strings attached, in plain language.
Let's talk →
Sources: LLM Stats · AI Tools Recap · dentro.de/ai · official announcements from Moonshot AI and Google.