Claude Fable 5.2 and Mythos 5.2 launch with broad, still-unverified capability claims

Anthropic presents Fable and restricted-access Mythos 5.2 as faster, cheaper upgrades spanning coding, reasoning, vision, science, hardware design, and agent workflows. Reports cite Fable at 49.1% versus 32.4% on an agentic coding benchmark, Mythos at 76.3% versus 61.7% on a cybersecurity benchmark, and a system-card SWE-bench Pro score of 93.0% for Fable; effort settings and company-selected testing limit comparisons. Repeated launch coverage and creative anecdotes do not substitute for independent capability and safety evaluation.