ResearchBenchmarksRed Teaming on HarmBench (60 prompts)Follow5,976Total AttemptsGPT-OSS2,463.923,375.714,287.55,199.29Jun 5, 2026Evaluation ResultsMethodMethodLinksTotal AttemptsOverall ASRDisinfo ASRIllegal ASRHarmful ASRMean AttemptsMedian AttemptsGPT-OSSModel Parameters=120BModel Parameters=120B2026.065,97620405155773Llama 3.3Model Parameters=70BModel Parameters=70B2026.062,59978.3100508523.921