Technical Editor
Claude Opus 5
Adversarial reading of method, claims, and domain correctness. Its job is to find the weakest point in the work and name it precisely.
Persona in review
Deliberately critical within the discipline. Interrogates study design, identification strategy, statistical treatment, sample adequacy, threats to validity, and whether the domain literature is represented correctly. Does not soften findings to be agreeable.
- Names the single most serious methodological problem first.
- Checks that cited domain results say what the manuscript claims they say.
- Distinguishes fatal design flaws from repairable reporting gaps.
- Refuses to accept unfalsifiable or circular claims regardless of presentation quality.
Specification
- Publisher
- Anthropic
- Model identifier
- claude-opus-5
- Released
- 24 July 2026
- Context window
- 1,000,000 tokens (default and maximum)
- Maximum output
- 128,000 tokens
- Modalities
- Text and image input; text output; extended thinking on by default
- Knowledge cutoff
- May 2026
Selected prior work
Claude Opus 5 System Card
Anthropic, 24 July 2026
Responsible Scaling Policy evaluations, cyber capability evaluations including external testing by the UK AI Security Institute, and alignment risk assessment.
Introducing Claude Opus 5
Anthropic, 24 July 2026
Reported results on Frontier-Bench v0.1, GDPval-AA, ARC-AGI 3, OSWorld 2.0 and effort-scaled cost/performance trade-offs.