Category: AI Research
-
China’s GLM-5.3 Can Now Build Its Own Cyberattacks
Anthropic put out a report that deserves more attention than it will probably get. Their security team tested an open model out of China called GLM-5.3, built by a company named Zhipu, also known as Z.ai. They wanted to know how good it had gotten at building real cyberattacks without any help. The results are…
-
Claude Spent 21 Hours and Found a New Piece of Biology
Anthropic’s scientists set a few hundred Claude agents loose on a database of roughly 200,000 enzymes, gave them 21 hours, and asked them to find something nobody had described before. They did. It’s called ART, short for array associated reverse transcriptases. It shows up in bacteriophages, the viruses that infect bacteria, and it looks like…
-
Anthropic Says Claude Now Leads a Quarter of Its Own R&D
Anthropic put out something that should stop you for a second. They built an internal scorecard, calling it the R&D Automation Index, and it says Claude now leads about a quarter of the actual work that goes into building the next Claude. Not helping with it. Leading it. That means Claude gets a high level…
-
One of ChatGPT’s Own Inventors Just Built Its Opposite
Diogo Almeida helped build the thing that made ChatGPT talk. He co-invented RLHF, the training method behind why ChatGPT sounds like a helpful assistant instead of a random text generator. Now he has a new company, and its first model does not talk at all. The model is called Jev, and it comes from his…
-
OpenAI Says Its AI Agents Cracked a 90-Year Math Problem
OpenAI says it just solved one of the six Millennium Prize problems, math puzzles so hard the Clay Institute has offered a million dollars each since the year 2000 and only one has ever been cracked before now. The claim: a swarm of about 10,000 of OpenAI’s own AI agents worked the Navier-Stokes existence and…
-
OpenAI’s AI Now Outworks Its Own Researchers 3 to 1
OpenAI just told everyone something wild. Inside its own research team, the AI agents are now putting in more hours than the humans. Roughly three times more. They call it hitting the automated research intern goal. Back in the fall they set a target: build a system that can take a research task, run with…
-
Tencent Just Open-Sourced a 770-Billion Model
Tencent just gave away one of the biggest AI models anyone has built, for free. The company open-sourced Hy4 preview, a model built on 770 billion parameters. It only lights up 49 billion of those for any given task, which is the trick that keeps it fast and affordable instead of impossibly expensive to run.…
-
The Internet’s Favorite Mystery AI Finally Has a Name
For almost a week, half the AI world was trying to figure out who built the mystery model everyone was suddenly using. It showed up on OpenRouter on August 20 with no company name attached, and it did not stay quiet for long. Within days it had climbed to the top of OpenRouter’s usage leaderboard,…
-

Anthropic Has a Smarter Model. You Won’t Get It
Anthropic just told the world it built something stronger than its own best public model, then told the world it is not letting anyone touch it. The company calls it Model 2. It beats Mythos 5, Anthropic’s current flagship, and there are no plans to ever release it. This came out of Anthropic’s August 2026…
-

Z.ai’s Coding Model Taught Itself to Chain Exploits
Z.ai set out to make its coding model better at finding software bugs. It ended up training something that can plan and carry out a full hack on its own, and that was not really the plan. GLM-5.3 uses the exact same base model as the previous version. Every gain came from more training on…
