By the Smili Phone team
Anthropic released two new Claude models on November 9, 2026, and the demo that came with them is hard to ignore. Claude Fable 5.2 and Claude Mythos 5.2 are built for difficult technical work: long coding projects and scientific research problems that earlier models struggled to finish. To show what they can do, an Anthropic engineer asked Fable 5.2 to build an operating system from scratch. The model coordinated more than 1,000 AI agents over eight days and produced a working system that could boot on a laptop, complete with windows, file storage, and a login screen.
That is a long way from the chatbots most of us use every day. A chatbot answers a question and stops. This experiment ran for more than a week, with agents working in parallel on different parts of the project, and the result actually ran on real hardware. It is the clearest sign yet of where the big AI labs are heading: models that manage complicated assignments over hours or even days, not just seconds.
One model, two doors
Fable 5.2 and Mythos 5.2 share the same underlying model, but they are not open to the same people. Fable 5.2 is available to everyone, with safeguards that block potentially dangerous cybersecurity and biology requests. Mythos 5.2 keeps those capabilities for verified organizations conducting approved research.
The split is a practical answer to a real problem. The same skills that help a researcher design a device could also help someone with harmful intent, so Anthropic draws a line between general access and restricted access. Other AI companies will face the same choice as their models get more capable.
More than a party trick
The operating system was not the only result Anthropic shared. The company also reported that the new models helped solve two previously open problems in black hole physics and designed three working electronic devices. Those tasks used to need a team of specialists and months of work.
The models also generate responses up to 30 percent faster than their predecessors. When an agent works on a task that takes hours, every speed gain compounds, and a faster model can try more approaches before it settles.
What it means for regular users
Most of us will never ask an AI to build an operating system. The useful part of this announcement is what it says about where everyday AI tools are going. The same ability to plan a long task, break it into pieces, and keep working without constant supervision is what makes an AI assistant useful for real work: drafting a long document, organizing a photo library, or troubleshooting a computer problem step by step.
It is worth keeping one caveat in mind. A single impressive experiment does not prove the model can handle every real engineering project on its own. The operating system worked, but it was a test run inside Anthropic, not a product shipping to customers. AI agents are getting much better at long, messy jobs, and the gap between a demo and a dependable tool is closing, but it has not closed yet.
If you want to try the new Claude yourself, start with a task that takes you an afternoon and see how far it gets before it needs your help. And if your own computer needs attention, a slow laptop, a dying battery, or a screen that stopped responding, bring it by our shop in Escondido and we will take a look together.
Related: AI Agents Score Only 12 Percent on Real Work · AI Agents Found 948 Ways Into Linux
