Claude Built a Working Operating System in 8 Days — Anthropic’s New Models Are HereTechnology

Anthropic also reported that the models helped solve two previously open problems in black hole physics and designed three working electronic devices, underlining the company’s push towards AI systems that can manage complicated assignments over hours or even days.

The models generate responses up to 30 per cent faster than their predecessors and cost less to run, according to the company. Fable 5.2 is priced at $8 per million input tokens and $40 per million output tokens.

Both models use the same underlying system, but access differs in an important way. Fable 5.2 is available to everyone, with safeguards that block potentially dangerous cybersecurity and biology requests. Mythos 5.2 retains those capabilities for verified organisations conducting approved research.

The launch comes as Anthropic makes a big bet on AI agents that can carry out long, multi-step work. The operating-system experiment offers a glimpse of that ambition, although it does not by itself establish that the model can independently handle every real-world engineering project.

The company has also been in the headlines for other reasons this week. Anthropic reportedly circulated details of a potential $69.3 billion initial public offering to bankers before retracting them, telling bankers the pricing information was inaccurate and that an updated filing would follow, according to Reuters.

Elsewhere in the AI industry, Fireworks AI disclosed multiple cyberattacks that exposed customer information and parts of unfinished AI models, while Nvidia reportedly tightened rules on how overseas partners rent its AI chips to Chinese customers. Meta also introduced Muse Ember, its smallest and most efficient model for repetitive AI-agent tasks, and Google unveiled two specialised AI chip systems, the TPU 8tx and 8i Max.