Claude Fable 5.2
ProductAm I affected?
Checked against NVD affected-version data on our server. Nothing you type is stored.
Recent news
Anthropic releases Claude Fable 5.2 and Claude Mythos 5.2 with stronger bug hunting
Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, which share an underlying model. The broadly available Fable applies stricter safeguards, with classifiers that block high-risk cybersecurity tasks such as penetration testing and exploit generation, and most biology research tasks. Blocked requests are answered by an Opus model instead, Opus 4.8 for cybersecurity and Opus 5 for biology.
Anthropic launches Claude Mythos 5.2 for finding and fixing software vulnerabilities
Anthropic has launched Claude Mythos 5.2, scoring 76.3% on CyScenarioBench and 89.6% on its internal CVE Rediscovery test, up from 61.7% and 78.3% for Mythos 5.1. Released alongside Claude Fable 5.2, it is gated through a Cyber Verification Program, with enterprise access also available via Claude Security. Anthropic says these benchmarks measure rediscovery of documented bugs, not zero-day detection rates.
Anthropic launches Fable 5.2 and Mythos 5.2 with lower prices and stronger vulnerability discovery
Anthropic has introduced Claude Fable 5.2 and Mythos 5.2, which share one underlying model with different safeguards, and cut API prices to $8 per million input tokens and $40 per million output tokens. Fable 5.2 scored 49.1% on the FrontierCode 2.0 benchmark against 32.4% for Fable 5.1, and Mythos 5.2 found 89.6% of vulnerabilities in an internal test of 1,829 previously disclosed flaws. Direct Mythos access is limited to vetted organisations, and analysts advise confirming zero-retention terms before using the models on sensitive workloads.
Anthropic releases Claude Fable 5.2 and Claude Mythos 5.2 AI models
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, which it says bring improvements in coding, reasoning and vision. The headline claims Claude Fable 5.2 outperformed GPT-6 Cyber on a multi-stage cyberattack benchmark, but the provided text is truncated and does not include the benchmark method or scores.