SOTA Security Evaluation on SecEval and PapersWithCode

92.3Accuracy

GPT-5

Updated 22d ago

Evaluation Results

Method
GPT-5 2026.01	92.3	0.002
GPT-4.1 2026.01	91.9	0.001
GPT-5-Mini 2026.01	91.1	0.003
O3-Mini 2026.01	90.8	0.002
Llama-3.3-70B-Instruct 2026.01	90.6	0.004
GPT-OSS-120B 2026.01	90.4	0.005
Phi-4 2026.01	89.8	0.004
Qwen-3-14B 2026.01	89.2	0.003
Qwen-3-8B 2026.01	88.6	0.003
GPT-5-Nano 2026.01	88.4	0.005
Llama-Primus-Nemotron-70B-Instruct 2026.01	88.2	0.005
GPT-OSS-20B 2026.01	87	0.006
Llama-3.3-70B-Ins-DAP 2025.06	86.4	-
Foundation-Sec-8B-Reasoning 2026.01	84.8	0.005
Llama-3.1-8B-Instruct 2026.01	83.2	0.003
Foundation-Sec-8B-Instruct 2026.01	82.9	0.003
Llama-Primus-Merged 2026.01	81.1	0.009
DeepHat-V1-7B 2026.01	79	0.005
DSR1D-Qwen-14B-DAP 2025.06	78.9	-
Llama-3.1-8B-DAP 2025.06	74.8	-
CyberPal-2.0-20B 2025.10	72.86	-
CyberPal-2.0-14B 2025.10	69.71	-
gpt-oss-120B 2025.10	68.02	-
gpt-oss-20B 2025.10	67.65	-
CyberPal-2.0-8B 2025.10	66.93	-
Llama-3-8B-Base 2025.06	62	-
Qwen3-14B 2025.10	61.48	-
CyberPal-2.0-4B 2025.10	59.02	-
Qwen3-4B 2025.10	57.38	-
Qwen3-8B 2025.10	56.19	-
Llama-Primus-Base (Cybersecurity specialized) 2025.06	50	-
Mistral-7B-v0.1 2025.06	43.7	-
Qwen-7B 2025.06	31.4	-