An
claude-opus-4-6
Benchmark & Insights
Anthropic Claude API
Updated Oct 4, 2026 All models
Sample size
232 runs
in window
Accuracy
75.3%
reference match · 251d
Confidence
90%
over 252 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 Leans low, calling 34 "MEDIUM_SAFE" days "low", but missed only 8 "unsafe" days, fewer than any newer Opus (4.8, 5, 5.5 miss 13-15).
- 02 Gets all 6 "HIGH" days "unsafe" with almost no false alarms.
- 03 Newer Opus versions have become more optimistic, not safer.
Recent forecasts