ResearchBenchmarksVulnerability Detection on 28 Real-world Logic Vulnerabilities Repositories (CVE dataset)Follow15True PositivesANTAEUS11.8812.6913.514.31Jul 1, 2026Evaluation ResultsMethodMethodLinksTrue PositivesFalse PositivesRecallANTAEUSReasoning Model=Opus 4...Reasoning Model=Opus 4.7, Number of Analyzed Functions (Fns)=48592026.07151,34515ANTAEUSReasoning Model=GPT-5....Reasoning Model=GPT-5.4, Number of Analyzed Functions (Fns)=48592026.07122,91812