Anthropic launches Claude Fable 5.2 and Mythos 5.2: Build operating system, design devices, and more

Anthropic’s latest AI models are being put to work on tasks that go well beyond answering questions, writing emails or generating code. The company says its upgraded Claude Fable 5.2 and Claude Mythos 5.2 can take on complex software development, scientific research, hardware design and cybersecurity work. In one experiment, the AI coordinated more than 1,000 agents to build an experimental operating system over eight days. It also designed electronic devices and worked on mathematical problems.

Alongside these capabilities, Anthropic says the new models respond up to 30% faster than their predecessors. However, not everyone will have access to both versions, as the company has introduced different restrictions depending on the type of work involved.

MUST READ | An AI model that can hack? OpenAI's new GPT-6 Cyber can find and exploit security flaws

What can Claude Fable 5.2 and Mythos 5.2 do?

One of Anthropic’s more notable demonstrations involved using Fable 5.2 to coordinate more than 1,000 AI agents over eight days to develop a working operating system prototype. The experimental system included drivers, storage, windows, a login screen and an installer.

The model also helped rewrite a software service from Python into Rust, a programming language often used for performance-intensive applications. Anthropic says the rewritten service started around 26 times faster.

Fable 5.2 was also used to design three working electronic devices: a home voice-assistant hub, a USB speakerphone and a handheld virtual pet. The designs received manufacturing support from Diode.

DON'T MISS | Claude can now check your bank transactions and spot unusual charges: Here’s how it works

Its scientific capabilities were also tested. Anthropic says 1,000 Fable agents converted an existing proof of Fermat’s Last Theorem into a computer-checkable mathematical format in 30 hours. The model also reportedly solved two open problems involving black holes and demonstrated improvements in robotics.

Mythos 5.2, meanwhile, was tested for cybersecurity and biological research. Anthropic says it identified more software vulnerabilities than earlier models and accurately predicted the structures of 28 out of 34 virus protein shells tested.

Claude Fable 5.2 and Mythos 5.2 - Pricing and availability

Both models generate responses up to 30% faster than their predecessors. Their standard pricing is $8 (around ₹750) per million input tokens and $40 (around ₹3,700) per million output tokens.

For faster responses, Anthropic offers an optional fast mode that increases token generation speed by up to 2.5 times. However, this comes at twice the standard price, taking the rates to $16 (roughly ₹1500) per million input tokens and $80 (roughly ₹7,500) per million output tokens.

Fable 5.2 is available through Claude platforms, Amazon Web Services, Google Cloud and Microsoft Azure. Mythos 5.2 has more restricted access and is limited to verified organisations working in areas such as cybersecurity and life sciences.

Anthropic also says it has reduced unnecessary safety blocks by around 55% compared with Fable 5.1 during early testing. The company has strengthened safeguards against attempts to extract its models’ capabilities through distillation.

For everyday coding and knowledge-based tasks, Anthropic continues to recommend Claude Opus 5.5, positioning Fable and Mythos for more demanding technical work.