← Back to directory
Anthropic
Claude Opus 4.5
Released Nov 24, 2025 · knowledge cutoff 2025-08
- Status
- Superseded
- Location
- United States
- Modality
- Multimodal
- Context window
- 200K
- Max output
- 64K
- Speed
- —
- Price ($/MTok in / out)
- $5 / $25
- Cost per task
- —
- Open weights
- No
Benchmarks4 of 8 reported
GPQA Diamond
87%
SWE-bench Verified
80.9%
LMArena Elo
1470
ARC-AGI-2
37.6%
MMLU-Pro
—
Terminal-Bench 2.1
—
AIME (latest)
—
Humanity's Last Exam
—
Strengths & weaknesses
Strengths
- Effort is a dial, so the same model does cheap quick work and slow deep work
- Unusually good at reading a large unfamiliar codebase before it changes anything
- Generalises to puzzle types it has plainly never seen rather than pattern-matching
Weaknesses
- Context felt tight for whole-repository work by the standards of its own moment
- Verbose by default — explains its reasoning even when you only wanted the answer
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor