An
claude-opus-5-5
Benchmark & Insights
Anthropic Claude API
Updated Sep 27, 2026 All models
Sample size
226 runs
in window
Accuracy
93.4%
consensus match · 226d
Confidence
83%
over 226 runs
Window end
Sep 27, 2026
most recent run
Input price
$4.00/MTok
prompt tokens
Output price
$20.00/MTok
completion tokens
Model insights
- 01 Slightly better than opus-5 and 20% cheaper, but it has the same blind spots: most of its miss days are the same days opus-5 missed.
- 02 It never overrates risk, but it said "safe" on several "unsafe" storm days, including Feb 18 and Feb 25.
Notes
Introduced on September 22, 2026 20% cheaper than Claude Opus 5. Has a breaking changes comparing to a previous version.
Recent forecasts