← Back to directory
Anthropic
Claude Opus 4
Released May 22, 2025 · knowledge cutoff 2025-03
- Status
- Superseded
- Location
- United States
- Modality
- Multimodal
- Context window
- 200K
- Max output
- 32K
- Speed
- —
- Price ($/MTok in / out)
- $15 / $75
- Cost per task
- —
- Open weights
- No
Benchmarks7 of 8 reported
MMLU-Pro
87.4%
GPQA Diamond
83.3%
SWE-bench Verified
72.5%
AIME (latest)
75.5%
Humanity's Last Exam
11.7%
LMArena Elo
1420
ARC-AGI-2
8.6%
Terminal-Bench 2.1
—
Strengths & weaknesses
Strengths
- Runs for hours on one task without losing the thread or looping
- Writes notes to itself and picks work back up across sessions
- Stops and asks when a requirement is ambiguous rather than guessing
Weaknesses
- Slow and deliberate — a poor fit for anything interactive
- Safety classifiers add refusals and latency to security and biology work
Released with Claude Sonnet 4 at the first Code with Claude conference.
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor