CWE-bench v1

Defensive vulnerability patching on 120 held-out tasks spanning 73 weakness types.

ModelsModels with a result.11
Top resultBest result on this test.68.0%GPT-6 Astra
Top-3 spreadPoints from first to third.0.0 pts
YearYear of release.2026

Result and price

020406080100$1$3$10$30$100RESULTOUTPUT PRICE PER 1M TOKENS · LOG SCALE

Results

11 results
#ModelResultThe score from the source.
01Grok 4.768.0%
06Grok 4.657.0%
10GPT-6 Sol52.0%
11Inkling37.0%