Anthropic

Claude Opus 4 (20250514)

Released 2025-05-22 · Data snapshot 2026-10-26

Overall score

56.5

Awaiting benchmarks (insufficient coverage)

Benchmark coverage

15.4%

4 third-party benchmarks

Context window

—

tokens

Input price

—

per million tokens

Output price

—

per million tokens

Dimension performance

Human preference

60.2

Knowledge

55.3

Above averageBelow averageAnchored at 50 (reference models’ average); every 15 points ≈ one standard deviation

Office1 items

Tax Eval v2nonthinking71.91

Knowledge2 items

MMLU-Prononthinking86.17

GPQA Diamond69.2

Human preference1 items

LMArena Human Blind Vote1414.176

Multiple reasoning tiers of the same benchmark take the best value; raw scores have different scales and are not comparable across benchmarks.