Anthropic's Claude Fable 5.2 Achieves 76.3% Success on CyScenarioBench Multi-Stage Cyber Operations
Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, reporting major gains in coding, reasoning, and vision capabilities. The models demonstrated the ability to complete complex software migrations, design electronic devices, advance black hole physics research, and formalize Fermat’s Last Theorem during training. On Irregular's CyScenarioBench, Fable 5.2 reached a 76.3% completion rate for realistic multi-stage cyber operations, surpassing both its predecessor and OpenAI's GPT-6 Cyber at 61%. Claude Mythos 5.2 is positioned as the company's strongest model for cybersecurity engineering, showing improved vulnerability detection, reproduction, and remediation. The release also includes updates to Claude Security for continuous code scanning and automatic fix tracking tied to project threat models.
The Hacker NewsNov 10, 2026Translated from English