Claude Mythos Preview is Anthropic's most capable model to date. Its most striking capability is in computer security — it can find and exploit software vulnerabilities at a level surpassing nearly all human experts, having discovered thousands of zero-day vulnerabilities in testing. Due to its cybersecurity potential, it is the first AI model restricted from public release. Access is limited to vetted partners via Project Glasswing.
Frontier-level reasoning capability
1M token context window
Exceptional at finding and exploiting software vulnerabilities
Follows complex instructions precisely
Broad factual knowledge base
Can autonomously use tools for security analysis
Moderate multilingual support
Capable but not specialized for creative tasks
Text-only model