An
claude-opus-5
Benchmark & Insights
Anthropic Claude API
Updated Oct 4, 2026 All models
Sample size
232 runs
in window
Accuracy
72.5%
reference match · 240d
Confidence
84%
over 240 runs
Window end
Oct 4, 2026
most recent run
Input price
$5.00/MTok
prompt tokens
Output price
$25.00/MTok
completion tokens
Model insights
- 01 Misses the most "unsafe" days of any Claude model, 15 of 25, calling 10 "MEDIUM_UNSAFE" days "medium" and "safe" and both late-September "HIGH" days "medium".
- 02 Never raises a false alarm.
- 03 Worse than the older claude-opus-4-6 at the same $5/$25.