Claude Opus 4.5
Claude Opus 4.5 cuts Opus pricing by two-thirds to $5/$25 and adds an effort parameter; Anthropic says it beat every human on its engineering take-home.
Price falls from $15/$75 to $5/$25 per million tokens. A new effort parameter lets it reach Sonnet 4.5-level performance at medium effort using 76% fewer output tokens. Over Sonnet 4.5 it gains 10.6% on Aider Polyglot, 29% on Vending-Bench and 15% on Terminal Bench.
- Date
- Monday, 24 November 2025
- Lab
- Anthropic
- Kind
- model
- Access
- closed API
- Price
- $5/$25 per M tokens in/out (down from $15/$75 for Opus 4/4.1)
Figures
| Measure | Value | Measured by |
|---|---|---|
| Output tokens at medium effort for Sonnet 4.5-level score | -76% effort parameter | company |
| Internal 2-hour performance-engineering take-home | above any human candidate Anthropic's own hiring exam; company claim | company |
API ID claude-opus-4-5-20251101. The SWE-bench Verified score is described only as highest among frontier models in the extracted text. Anthropic's Apr 2026 Managed Agents post says Opus 4.5 removed 'context anxiety' in long agent runs.
Sources
This record was checked against its sources on 6 October 2026. How we check