Anthropic is reportedly testing a new AI model, Claude Mythos, which promises significant advancements over existing products. According to PANews, the model has been made available to a select group of early clients. Internal documents reveal that the model, codenamed 'Capybara,' surpasses the capabilities of the previous Claude Opus 4.6, showing marked improvements in software programming, academic reasoning, and cybersecurity testing.
Anthropic has accumulated nearly 3,000 unpublished assets in a private data cache, including blog drafts about Claude Mythos/Capybara and materials from a closed-door summit for European CEOs. The company claims the model excels in cybersecurity defense, outperforming other AI systems. Due to high operational costs and associated risks, Anthropic plans to initially target cybersecurity defenders with the model rather than offering it for widespread use.