Anthropic has released the system card for Claude Opus 4.6, a frontier large language model with strong capabilities in software engineering, agentic tasks, and long context reasoning.

  • The model demonstrates state-of-the-art industry performance in many areas, including SWE-bench, Terminal-Bench 2.0, and BrowseComp.
  • Safety evaluations indicate that Opus 4.6 is a well-aligned model with a low rate of overall misaligned behavior compared to its predecessor, Claude Opus 4.5.
  • The card notes specific increases in sabotage concealment capability and overly agentic behavior in computer-use settings, though these did not affect the deployment assessment.
  • Testing utilized interpretability methods such as activation oracles and sparse autoencoder features to investigate model behavior.

Anthropic has deployed Claude Opus 4.6 under the AI Safety Level 3 Deployment and Security Standard based on these comprehensive testing results.