Anthropic introduces Opus 5.5 with stronger judgment and lower prices
The first Claude 5.5 release targets complex work and includes new behavioral evaluations and safeguards.
THE STORY IN THREE POINTS
- Opus 5.5 opened the Claude 5.5 family on September 22.
- The launch reports lower running costs than Opus 5.
- Anthropic expanded tests for boundaries, long tasks and impossible assignments.
The release
Anthropic announced Claude Opus 5.5, the first model in the Claude 5.5 family. The company describes better performance on complex work and reports typical default-setting costs 40% below Opus 5.
The launch also discusses external evaluation and behavioral tests covering longer assignments, impossible tasks and boundaries on actions. Anthropic reports improved results, while acknowledging remaining limitations. We have not reproduced the company’s performance or safety measurements.
Editorial context: judgment is visible in exceptions
A demanding assignment should test what happens when an input is missing or an objective cannot be satisfied. A dependable collaborator would explain the limitation and preserve the owner's intent rather than quietly substitute an easier task.
Reviewers could therefore include deliberately incomplete instructions in their evaluation. The goal would be to observe whether the model asks a useful question, continues independent work where appropriate and makes its remaining uncertainty understandable.
Editorial outlook: what to watch
Evidence from actual deployments should connect model behavior with the quality of the completed work.
Useful reports would describe failed attempts, review requirements and the conditions under which people intervened. That would make a claim about improved judgment more concrete than a benchmark margin alone.
Sources: Anthropic: Introducing Claude Opus 5.5 ↗
