report · 2026-09-22
Summary of METR's predeployment evaluation of Claude Opus 5.5
On this page
Why it matters here
METR's preliminary September 22 assessment reports modest AI research capability improvements for Claude Opus 5.5 relative to Fable 5.1 while judging full AI R&D automation unlikely. It draws on five targeted tasks tested over 10 business days, an Anthropic questionnaire and researcher interview, and limited-access background assessments. METR also judged the model's development somewhat, but probably not dramatically, accelerated by AI. These are qualified evaluator conclusions rather than new benchmark scores or demonstrated recursive improvement.
AI-R&D capabilityEvaluation integrity
What to keep in mind
- Preliminary testing over 10 business days on five selected tasks; complete run cohorts, budgets, model configuration and raw results are not established by this summary.
- METR worked under an unpaid assessment agreement; Anthropic could review and edit the initial summary and METR signed off on the final text. This is not unrestricted independent access or experimental replication.
- Questionnaire answers, a researcher interview, conclusions from a separate METR team and an undisclosed background source also inform the assessment. The separate team's supporting evidence was unavailable to the authors of this summary.
- The quoted approximately 1.5× acceleration estimate comes from a highly preliminary separate report whose time period is unspecified. It is not adopted as a checked quantitative observation.
- The report's comparative conclusions about Fable 5.1 are attributed judgments, not a new directly comparable benchmark score or proof of statistical separation.
- The report does not assess Anthropic policy-threshold compliance or alignment properties, demonstrate autonomous recursive self-improvement, or distinguish accelerating from steady capability growth.
- Source publication is September 22, 2026; this is newly checked historical source context, not an October 14 model release. Related system-card evidence is already tracked separately.