Specification Alignment on SPECBENCH Average over scenarios benchmark leaderboard · SOTA2 Research

Specification Alignment on SPECBENCH Average over scenarios

93.73Safety ScoreGPT-5-chat

Evaluation Results

Method	Links
GPT-5-chat2025.09		93.73	82.54	82.14
STAIR-Llama-3.1-8B-DPO-32025.09		89.27	36.86	48.25
Gemini-2.5-flash-thinking2025.09		84.73	73.49	68.59
RealSafe-R1-8B2025.09		83.6	37.37	45.93
Gemini-2.5-flash-lite-thinking2025.09		80	73.59	65.02
Gemini-2.5-pro2025.09		79.6	73.89	65.03
Llama-3.2-1B-Instruct2025.09		79.33	27.3	37.61
GPT-4.12025.09		78.87	82.69	69.2
DeepSeek-R12025.09		78.6	78.94	67.12
Qwen3-32B-thinking2025.09		71.8	75.98	60.12
GPT-4.1-mini2025.09		71.4	72.38	57.24
Gemini-2.5-flash2025.09		71.27	71.45	55.85
DeepSeek-V32025.09		70.13	72.99	57.65
Qwen3-14B-thinking2025.09		70	73.76	57.32
Qwen3-32B2025.09		68.67	67.73	52.47
DeepSeek-R1-Distill-Llama-70B2025.09		68.53	55.32	47.8
Gemini-2.5-flash-lite2025.09		68.33	63.8	50.15
Llama-3.3-70B-Instruct2025.09		68.07	69.38	53.8
Qwen3-8B-thinking2025.09		66.73	71.76	53.83
Qwen3-14B2025.09		64.27	70.58	51.03
Llama-3.2-3B-Instruct2025.09		63.67	47.65	40.49
Qwen3-4B-thinking2025.09		63.47	68.04	50.26
Qwen3-8B2025.09		62.53	66.88	48.58
Qwen3-4B2025.09		59.13	59.54	43.33
DeepSeek-R1-Distill-Qwen-32B2025.09		57.8	52.82	39.53
Llama-3.1-8B-Instruct2025.09		56.87	65.99	44.54
Qwen3-0.6B2025.09		56.47	31.85	30.74
Qwen3-1.7B2025.09		55.27	47.48	36.34
Mistral-Small-Instruct-24092025.09		53.67	59.32	40.42
DeepSeek-R1-Distill-Llama-8B2025.09		53.67	45.44	35.01
Qwen3-0.6B-thinking2025.09		52.8	41.15	33.59
Qwen3-1.7B-thinking2025.09		52.6	54.81	37.8
Mistral-7B-Instruct-v0.32025.09		51.27	51.21	36.52