$ ls benchmarks/ --cards all --count 59

Every benchmark in
Claude's current system cards

All 59 named, citable benchmarks reported across Anthropic's four most recent Claude system cards — Fable 5.1 & Mythos 5.1, Opus 5, Sonnet 5, and Haiku 4.5. Each one is broken down the same five ways: what it's for, how it was built, how it's run, how it's scored, and how to read the score — every claim traced to the card's own text, and every gap in that text named rather than filled in.

59benchmarks
13evaluation areas
4system cards
11safety evals
Group by
Card Type
Search