Anthropic has published the system card for Claude Opus 5.5, describing it as their most powerful model by some measures and claiming parity with or superiority over Fable 5.1 while being cheaper than Opus 5.
- The model is classified as CB-1 capable for chemical and biological weapons but not CB-2, utilizing safeguards similar to Fable 5.1.
- Anthropic has shifted from testing helpful-only versions to refusal-avoiding tests to better evaluate dual-use capabilities.
- Opus 5.5 achieved a new high in black-box RNA sequence design, performing competitively with top US labor-market performers on medium-horizon tasks.
- The system card outlines specific classifier fallbacks, such as using Opus 4.8 for cyber misuse and blocking distillation attacks entirely.
The document provides a comprehensive overview of the model's safety thresholds, evaluation methodologies, and performance benchmarks relative to previous iterations.