According to Forbes, Anthropic has acknowledged testing a new model called Claude Mythos, which offers a "quantum leap" improvement over existing products, and has opened it to a small group of early customers. Leaked internal drafts reveal that the model, internally codenamed "Capybara," significantly outperforms the previously strongest Claude Opus 4.6 in software programming, academic reasoning, and cybersecurity testing. Anthropic has left nearly 3,000 unpublished assets in its private data cache, including draft blog posts introducing Claude Mythos/Capybara and materials from a European CEO invitation-only closed-door summit. The company states that the model "far exceeds other AIs" in cyber offense and defense, plans to initially target cybersecurity defense teams, and has not yet fully opened it due to high operational costs and significant risks.