Anthropic to launch new 'Claude Mythos' model with advanced reasoning features

Anthropic PBC confirmed in a Thursday statement to Fortune that engineers have finished training a new large language model called Claude Mythos. The company is piloting the model with early customers. It described the LLM as the most capable it has built to date. A configuration error exposed a content management system folder containing 3,000 assets. The folder included a draft launch blog post for Claude Mythos. The document showed that the model will cost more than the company's existing algorithms. Anthropic intends to add a fourth product tier for the new model. The tier, referred to as Capybara in the draft, will exceed the capabilities of the current Opus version of Claude 4.6. Internal tests indicated that this edition outperforms Claude 4.6 Opus on programming tasks and other reasoning applications. The model demonstrates particular strength in detecting cybersecurity vulnerabilities. Anthropic plans precautions to block unauthorized access by hackers. The draft post stated that the model indicates an upcoming wave of systems able to exploit vulnerabilities faster than defenders can address them. The company will provide early access to organizations. Recipients can use the opportunity to enhance protections for their codebases against AI powered exploits. Anthropic launched Claude Code Security last month, a tool powered by Claude Opus 4.6 that has uncovered more than 500 high severity exploits in open source projects. Shares of CrowdStrike Holdings Inc. and Palo Alto Networks Inc. declined more than 5 percent on the disclosure. Market participants cited worries about competition in vulnerability detection.