Benchmarks
Compare systems
Pick up to five systems from one evaluation and compare their results.
Side by side3 of 3 systems on the Memory safety · Exploit versus repair leaderboard
Each system has two bars on one set of vulnerabilities: the share it exploited, and the share it repaired. This is a paired view of the April 2026 vulnerability repair data. The reasoning effort for these runs is not recorded.
SystemExploit · Repair
Claude Code Opus 4.7
79.9% · 57.4%repair minus exploit -22.5 pp, -30.8 to -13.9209 vulnerabilities
OpenCode GPT-5.5
96.2% · 85.2%repair minus exploit -11.0 pp, -16.5 to -6.2209 vulnerabilities
OpenCode GPT-5.3 Codex
77.0% · 83.7%repair minus exploit +6.7 pp, -0.3 to +13.8209 vulnerabilities
Leaderboard: Memory safety, Exploit versus repair· this leaderboard is not ranked · Open the leaderboard