Anthropic Readies Public Release of Its Most Powerful AI Model Yet

Anthropic is preparing to release its most advanced generative AI model to the general public, after initially restricting access to a select group of partners. The model, known internally as Claude Mythos Preview, has been part of a controlled rollout under Project Glasswing, where cybersecurity experts and major tech companies have been testing its ability to identify software vulnerabilities.

In a blog post on Thursday, Anthropic confirmed that it is making “swift progress” on the safety measures needed before Mythos-class models can be broadly deployed. The company expects to make these models available to all customers within weeks.

The staged release has sparked debate. Some security researchers view it as responsible stewardship of a powerful tool that can find exploits faster than human hackers—Mozilla’s latest Firefox update, for instance, included over 200 fixes identified by Mythos Preview. Others see a marketing strategy that builds anticipation.

Darren Williams, CEO of BlackFog, noted that powerful model launches are always high-stakes moments. “The window between release and widespread defense adoption is a vulnerable time,” he said.

Cost could temper adoption. Cybersecurity researcher Jake Williams of IANS Research said Mythos is roughly 30 times more expensive to run than Anthropic’s previous Opus model, putting it out of reach for most threat actors. “Nation-state actors already had better tools,” he added.

Meanwhile, the company released Claude Opus 4.8, a modest upgrade over Opus 4.7 with improved accuracy and reduced unsupported claims. New features allow users to adjust how much processing power the model uses for each response, balancing speed against thoroughness.