Category: AI Research
-

Alibaba’s New Model Doesn’t Need the Cloud at All
Alibaba just put a 27 billion parameter model out in the open, and the whole point of it is that you do not need a data center to run it. Qwen3.8-27B ships under Apache 2.0, so anyone can download it, fine tune it, sell products built on it, whatever they want. It reads text, images,…
-

Anthropic Set Three Claude Agents Loose. They Went to War
Anthropic ran an experiment that sounds almost too weird to be real. They took three versions of Claude, put them all to work on the exact same software project, gave each one instructions that quietly clashed with the others, and never told any of them the other two existed. What happened next is the part…
-

An AI Just Designed 16 Working Viruses From Scratch
Researchers at Stanford and the Arc Institute trained an AI model called Evo 2 on huge libraries of DNA, the same way language models are trained on text. Then they asked it to design something that has never existed before, a complete viral genome, built from nothing. They tested nearly 300 of the AI’s designs…
-

OpenAI’s Next Model Just Cracked Ten Old Math Problems
OpenAI put out something quiet but genuinely wild this week. An internal version of its next major model, codenamed Astra, solved ten math and theoretical computer science problems that nobody had made real progress on in at least a decade, and in some cases much longer. These are not homework problems. We’re talking high dimensional…
-

Google DeepMind Teaches a Robot to Use Its Whole Body
Most robot demos you’ve seen are an arm on a table, picking up a block, over and over. Google DeepMind just showed something different: a humanoid that walks across a room, bends its knees, and places an object on a shelf, all as one continuous motion instead of a chain of separate tricks. The system…
-

Google Just Broke Up Its Nobel-Winning AI Team
Here is a strange one. The team that built AlphaFold, the AI system that won Google DeepMind an actual Nobel Prize, has basically been taken apart. The Financial Times reported that most of the original AlphaFold paper authors have been moved onto other projects over the past year. Some went to work on Gemini directly.…
-

OpenAI Just Opened Its Best Models to Scientists
OpenAI just made a pretty specific promise. Give 100,000 academic researchers free access to its most capable AI models, and get out of the way while they use it. The program starts small, 10,000 researchers this summer, then grows toward that 100,000 mark by 2027. Qualifying faculty and postdocs get GPT-5.6 Sol Pro, the company’s…
-

Google Just Mapped 15 Million AI Conversations
Google just published the first real map of how people actually use AI, and the answer isn’t quite what most of the doom headlines have been telling you. The report is called the AI and Economy ATLAS, and it’s built from 15 million anonymized conversations pulled from the Gemini app, AI Mode in search, and…
-

Anthropic Just Made Its Best AI Model Half Price
Anthropic just put out Opus 5, and the headline isn’t that it’s smarter. It’s that it’s half the price of what came before it, and in a bunch of tests it actually beats its own bigger, pricier sibling, Fable 5. That’s the part worth sitting with. For a while the story in AI was simple:…
-

Google DeepMind Just Put Brains in Boston Dynamics Robots
Boston Dynamics is now running Gemini Robotics integrated into its Spot robot. That partnership, announced this week through Google Cloud, means the robots that navigate physical space and make real decisions are now powered by multimodal AI that understands what it sees and what to do about it. The Spot robot with Gemini is being…
