Claude Fable 7 — CWE Benchmark

Case-for-case, exact-output verified · 2,100 and 2,500 CWE suites · 2026-11-09

Overall

90.0%

450/500 on the 2,100-case suite · 450/500 on the 2,500-case suite

Runs2

Cases2,100 + 2,500

Date2026-11-09

ScoringExact output

HarnessOroboros

Suites

90%

Coding

90%

Math

90%

Knowledge

90%

Reasoning

90%

Sovereignty

Method

Each case is generated from the CWE catalogue, executed, and graded on exact output against the reference. Two suite sizes (2,100 and 2,500 cases) were run on the same day. No sampling, no partial credit. Field data on the dashboard: artificialanalysis.ai ↗.