Security Evaluation on SecEval benchmark leaderboard · SOTA2 Research

Security Evaluation on SecEval

92.3AccuracyGPT-5

Evaluation Results

Method
GPT-52026.01	92.3	0.002
GPT-4.12026.01	91.9	0.001
GPT-5-Mini2026.01	91.1	0.003
O3-Mini2026.01	90.8	0.002
Llama-3.3-70B-Instruct2026.01	90.6	0.004
GPT-OSS-120B2026.01	90.4	0.005
Phi-42026.01	89.8	0.004
Qwen-3-14B2026.01	89.2	0.003
Qwen-3-8B2026.01	88.6	0.003
GPT-5-Nano2026.01	88.4	0.005
Llama-Primus-Nemotron-70B-Instruct2026.01	88.2	0.005
GPT-OSS-20B2026.01	87	0.006
Llama-3.3-70B-Ins-DAP2025.06	86.4	—
Foundation-Sec-8B-Reasoning2026.01	84.8	0.005
Llama-3.1-8B-Instruct2026.01	83.2	0.003
Foundation-Sec-8B-Instruct2026.01	82.9	0.003
Llama-Primus-Merged2026.01	81.1	0.009
DeepHat-V1-7B2026.01	79	0.005
DSR1D-Qwen-14B-DAP2025.06	78.9	—
Llama-3.1-8B-DAP2025.06	74.8	—
CyberPal-2.0-20B2025.10	72.86	—
CyberPal-2.0-14B2025.10	69.71	—
gpt-oss-120B2025.10	68.02	—
gpt-oss-20B2025.10	67.65	—
CyberPal-2.0-8B2025.10	66.93	—
Llama-3-8B-Base2025.06	62	—
Qwen3-14B2025.10	61.48	—
CyberPal-2.0-4B2025.10	59.02	—
Qwen3-4B2025.10	57.38	—
Qwen3-8B2025.10	56.19	—
Llama-Primus-Base (Cybersecurity specialized)2025.06	50	—
Mistral-7B-v0.12025.06	43.7	—
Qwen-7B2025.06	31.4	—