Hey there! Welcome to your weekly AI round-up!
This week, things are moving fast: OpenAI’s o3-Pro is already shaking up the market with rock-bottom prices, Mistral’s making big open-source waves, and Meta’s robots are finally learning how to “read the room”—literally. Google’s tackling cyclones with DeepMind, Mistral dropped an entire compute platform, and Hollywood just declared war on AI image generators. Oh, and toy giant Mattel? They’re making Barbie smarter than ever.
Let’s get into it!
This Creators’ AI Edition:
Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
New AI Agent Use Cases: Software Testing AI Agents
TestSprite just leveled up. With the new Model Context Protocol, it plugs right into your favorite code editors and auto-generates integration tests, supervising your AI coding agents like a pro. Enjoy faster execution, better accuracy, and multi-provider AI support. Now with automated test monitoring, scheduling, and a sleek new UI that even non-tech folks can use.
Test smarter. Ship faster. Try the free community version now.
Featured Materials 🎟️
OpenAI slashes prices as it releases the new o3-pro model
OpenAI has just rolled out o3-Pro, its latest and most advanced reasoning model, and it’s making waves not just for its intelligence but for its price tag. Designed to “think longer,” o3-Pro is now the top performer on tough benchmarks in math, science, and programming, with testers saying it feels way smarter than earlier models and stands out from the competition. It can handle tasks like web search and data analysis, though it’s a bit slower and doesn’t support image generation or Canvas yet. ChatGPT Pro and Team users get access right away, with Enterprise and Edu customers following soon.
The real headline? OpenAI has slashed o3-Pro’s prices by a huge 80 percent, making this leap in capability shockingly affordable and putting direct pressure on rivals like Google and Anthropic.
Mistral released Magistral
Mistral AI just launched Magistral, their new reasoning model designed to make complex thinking much more approachable. It comes in two versions: Magistral Small, which is open-source and has 24 billion parameters, and Magistral Medium, which is geared toward enterprise users looking for top performance. Magistral is built to handle structured calculations, logical reasoning, decision trees, and rule-based systems, all while supporting multilingual use and clear, step-by-step thought processes. Magistral Medium offers quick responses with features like Think Mode and Flash Answers, while Magistral Small is open for the community to explore and build on.
News of the week 🌍
Meta unveils AI that understands real-world physics
Meta has introduced V-JEPA 2, a powerful “world model” that gives AI a much better understanding of physics and real-world outcomes, helping robots handle objects they’ve never seen and navigate unfamiliar spaces. Trained on over a million hours of video, the 1.2-billion parameter model can predict how things move and interact, and it managed 65 to 80 percent success rates in picking and placing new objects during tests. Meta says V-JEPA 2 is also incredibly fast, running 30 times quicker than Nvidia’s Cosmos model while topping video understanding benchmarks. Alongside the release, Meta unveiled new tests showing humans are still better at physical reasoning, but their model closes the gap. This is a big step for AI, helping it move beyond text and actually deal with the messiness of the real world, making robots and AI agents more adaptable and useful in everyday situations.
Google DeepMind launched Weather Lab
Google DeepMind and Google Research have just launched Weather Lab, a new interactive site where anyone can explore Google’s latest AI-powered weather models, including their newest experimental system for predicting cyclones. The technology is still in early stages, but the aim is to help weather agencies and emergency teams improve their ability to predict the path and intensity of cyclones. For example, their model recently showed impressive accuracy by predicting the routes of several cyclones near Madagascar and in the Indian Ocean up to a week ahead. Google is also inviting feedback from experts to help make these models even better.
Meta Invests $15B in Scale AI, Hires Alexandr Wang
Meta is making a huge move in the AI world by spending nearly $15 billion to buy a 49 percent stake in Scale AI, valuing the data-labeling company at about $29 billion, and bringing Scale’s 28-year-old CEO, Alexandr Wang, into the company as a top AI executive. Wang, an MIT dropout who became the world’s youngest self-made billionaire, will lead Meta’s new superintelligence team and help strengthen its AI data partnerships. This deal is both a strategic investment and a major talent grab, as Meta looks to regain its edge in generative and general AI after staff changes and delays with its big LLaMA and Behemoth models. The partnership could also shake up the industry, since Scale works with many other AI clients, and may bring new regulatory questions as Meta doubles down on its AI ambitions.
Mistral launched Mistral Compute
Mistral AI just introduced Mistral Compute, a new platform that gives you direct access to powerful AI infrastructure, from top-notch GPUs to easy APIs and orchestration tools. Launched in June 2025, Mistral Compute is all about giving businesses and organizations the freedom to build, train, and run their own AI models without relying on the big cloud providers. It’s designed with a Europe-first mindset, putting a big focus on privacy, data sovereignty, and sustainability by using low-carbon energy. Early partners like BNP Paribas, Orange, and Thales are already on board, showing just how versatile and enterprise-ready this new AI stack really is.
ByteDance released Seedance
ByteDance, the company behind TikTok, has just introduced SeedDance, a new video generation model they say outperforms Google’s Veo-3. Along with the model, ByteDance also launched a fresh video benchmark and an online arena for testing and comparing video generation tools, making it easier to see how SeedDance stacks up against the competition.
Disney and Universal file copyright lawsuit against Midjourney
Disney, Universal, and other Hollywood heavyweights have just launched a major lawsuit against AI image generator Midjourney, accusing the startup of turning their famous characters into an endless stream of copyright violations. This is the first big legal move from the studios against an AI company, and the complaint includes plenty of examples, with side-by-side images showing lookalikes of Yoda, Shrek, Spider-Man, and Minions. Disney’s lawyers say they still see huge potential in AI, but they draw a hard line at what they call piracy. While some media outlets have already made deals with AI companies, Hollywood is taking a tougher stance, and the outcome of this case could set important legal ground rules for how AI uses copyrighted content.
OpenAI teams up with Mattel to create AI-powered toys
OpenAI and Mattel have teamed up in a new partnership that will bring AI-powered toys and experiences to classic brands like Barbie, Hot Wheels, and American Girl. The collaboration means OpenAI’s technology will be woven into both physical toys and digital games, with the first AI-enhanced product arriving later this year. Mattel employees are also getting access to ChatGPT Enterprise to help spark creativity and streamline their work. Both companies are putting a strong focus on safety and making sure everything is age-appropriate, with Mattel keeping full control over its brands and products. This move promises to make toys smarter and more interactive, but it’s also sure to spark debate about privacy and whether AI belongs in the hands of kids.
Useful tools ⚒️
FuseBase AI Agents - Internal and external workspaces powered by AI agents
Dia Browser - The AI browser where you can chat with your tabs
Commerce Copilot -Your Personalised AI Assistant
Instance - From idea to app–in an Instance (iOS, Android & Web)
Hugging Face AISheets -AI-powered spreadsheets for everyone
AISheets is a new open-source tool from Hugging Face that brings thousands of AI models into a simple spreadsheet interface. Easily process, analyze, and automate your data with AI.
Weekly Guides 📕
How to prompt Veo 3 for the best results
How Two Engineers Ship Like a Team of 15
Adapting SEO for LLMs and AI search by Vercel
Build an Agentic RAG App with Reasoning
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
OpenAI taps Google in unprecedented cloud deal
First Zero-Click AI Vulnerability Enabling Data Exfiltration from Microsoft
What do you think about Meta Acquiring Scale for $15B? Are they gonna kill the #1 AI labeling company by that?
Meta just acquired 49% of Scale AI.
Alexandr Wang stepped down to lead “Superintelligence” inside Meta.
But is this about innovation… or silent control?
👉 https://substack.com/@marcokindermann/note/c-125713768
#Meta #ScaleAI #Superintelligence #WeAreNotData #AIethics