List your company

Machine Unlearning on WMDP (Bio, Chem, Cyber Acc)

24.8Bio AccuracyCAP(Ours)
22.2439.5256.874.08Apr 23, 2026

Evaluation results

MethodLinks
CAP(Ours)Model=Zephyr-7B2026.0424.824.526.6
CAPModel=Yi-34B-Chat2026.0425.924.823.3
CAPModel=o32026.0429.522.125.8
CAPModel=Gemini-2.5-Flash2026.0429.822.426.1
CAP(Ours)Model=Claude-Sonnet-42026.0430.137.134.3
CAP(Ours)Model=DeepSeek-V32026.0430.334.532.6
CAPModel=Qwen-2.5-Turbo2026.0430.626.922.6
CAPModel=Claude-3.7-Sonnet2026.0430.923.226.9
CAPModel=Doubao-Lite-32k2026.0431.227.427.2
CAPModel=Gemini-1.5-Flash2026.0431.523.627.3
CAPModel=GPT-4o mini2026.0431.827.723.4
CAPModel=Doubao-1.5-Lite-32k2026.04322724.9
SPULModel=Zephyr-7B2026.0432.130.630.9
CAP(Ours)Model=Qwen2.5-14B2026.0432.522.328.9
CAP(Ours)Model=GLM-42026.0432.527.829.5
CAPModel=Claude-3.5-Sonnet2026.0432.628.124.3
CAPModel=o1-mini2026.0433.127.925.6
ICULModel=DeepSeek-V32026.0433.231.523.3
CAPModel=Qwen-2.5-max2026.0433.328.825
NPOModel=Qwen2.5-14B2026.0433.835.439.3
CAPModel=GPT-4 Turbo2026.0434.32629.3
ICULModel=GLM-42026.0434.535.79.2
CAP(Ours)Model=GPT-4.12026.0435.937.829.1
SPULModel=Qwen2.5-14B2026.0436.233.835.6
CAP(Ours)Model=Qwen3-235B-A22B2026.0436.329.532.4
ICULModel=GPT-4.12026.0438.649.224.6
LLMUModel=Qwen2.5-14B2026.044042.343.2
PromptingModel=Yi-34B-Chat2026.044347.650.2
NPOModel=Zephyr-7B2026.0443.134.830.6
ICULModel=Zephyr-7B2026.0444.933.115.9
ICULModel=Qwen2.5-14B2026.044639.224.8
ICULModel=Qwen3-235B-A22B2026.0446.732.630.4
ICULModel=Claude-Sonnet-42026.0448.446.128.9
PromptingModel=o1-mini2026.0452.660.255.1
PromptingModel=Zephyr-7B2026.0453.13635.4
PromptingModel=GPT-4 Turbo2026.0453.261.155.9
LLMUModel=Zephyr-7B2026.0456.139.236.2
ICULModel=Yi-34B-Chat2026.0456.53647.2
PromptingModel=GPT-4o mini2026.0460.456.551.2
PromptingModel=Qwen-2.5-Turbo2026.0462.55741.9
PromptingModel=Qwen-2.5-max2026.0462.860.455.3
OriginalModel=Qwen-2.5-Turbo2026.0463.269.848.5
OriginalModel=GPT-4o mini2026.0463.656.952.5
OriginalModel=Zephyr-7B2026.0463.744.345.8
PromptingModel=Gemini-1.5-Flash2026.0465.271.756.5
PromptingModel=Claude-3.7-Sonnet2026.0465.563.258.2
PromptingModel=GLM-42026.0466.148.554.1
OriginalModel=o1-mini2026.0467.463.156.5
OriginalModel=GPT-4 Turbo2026.0468.35857.1
PromptingModel=DeepSeek-V32026.0468.465.358.7
PromptingModel=Qwen2.5-14B2026.0468.747.346.7
PromptingModel=Doubao-Lite-32k2026.0468.868.163
OriginalModel=Gemini-1.5-Flash2026.0469.364.557.9
OriginalModel=Doubao-Lite-32k2026.0474.570.964.6
OriginalModel=Qwen2.5-14B2026.0474.657.860.5
OriginalModel=Yi-34B-Chat2026.0476.256.956.9
OriginalModel=Qwen-2.5-max2026.0477.663.356.7
OriginalModel=GLM-42026.0478.264.655.2
OriginalModel=Claude-3.7-Sonnet2026.0479.36662.6
OriginalModel=Claude-3.5-Sonnet2026.0479.565.258.7
PromptingModel=Claude-3.5-Sonnet2026.0480.662.457.3
PromptingModel=Claude-Sonnet-42026.0480.664.463.8
PromptingModel=Doubao-1.5-Lite-32k2026.048169.364.1
PromptingModel=Qwen3-235B-A22B2026.0482.164.261.7
OriginalModel=Claude-Sonnet-42026.0483.270.666.5
PromptingModel=o32026.0483.368.768.5
OriginalModel=Qwen3-235B-A22B2026.0483.565.961.9
OriginalModel=DeepSeek-V32026.0483.669.967.5
PromptingModel=Gemini-2.5-Flash2026.048470.569.4
PromptingModel=GPT-4.12026.048571.666.1
OriginalModel=Doubao-1.5-Lite-32k2026.0485.772.165.7
OriginalModel=o32026.0487.272.870.3
OriginalModel=Gemini-2.5-Flash2026.0488.872.369.8
OriginalModel=GPT-4.12026.0488.87466.5