List your company

General Capability on MMLU

79.6MMLU AccuracyOriginal Weight Model
44.2453.4262.671.78Apr 2, 2025Jun 8, 2025Aug 15, 2025Oct 21, 2025Dec 28, 2025Mar 5, 2026May 12, 2026

Evaluation results

MethodLinks
Original Weight ModelModel=Qwen2.5 14B Instruct2025.0479.6
REPBENDModel=Qwen2.5 14B Instruct2025.0478.89
QWEN 3 14BModel=QWEN 3 14B2025.1076.1
ELSModel=QWEN 3 14B2025.1076.1
SurgicalModel=QWEN 3 14B2025.1074.7
AdaSteerModel=QWEN 3 14B2025.1074.4
CASTModel=QWEN 3 14B2025.1074
AlphaSteerModel=QWEN 3 14B2025.1073.3
QWEN 3 8BModel=QWEN 3 8B2025.1072.8
ELSModel=QWEN 3 8B2025.1072.8
AlphaSteerModel=QWEN 3 8B2025.1072.3
SurgicalModel=QWEN 3 8B2025.1071.2
AdaSteerModel=QWEN 3 8B2025.1070.9
CASTModel=QWEN 3 8B2025.1070.5
system promptModel=QWEN 3 8B2025.1070.2
system promptModel=QWEN 3 14B2025.1069.8
LLAMA3.1-8B-INSTModel=LLAMA3.1-8B-INST2025.1068.1
ELSModel=LLAMA3.1-8B-INST2025.1068.1
Llama-3.1-8B-ITModel Status=Base Model, Evaluation Metric Type=Acc2025.1068
ELSBase Model=Llama-3.1-8B-IT, Evaluation Metric Type=Acc2025.1068
SurgicalModel=LLAMA3.1-8B-INST2025.1067.7
CASTModel=LLAMA3.1-8B-INST2025.1067.3
AlphaSteerModel=LLAMA3.1-8B-INST2025.1066.7
WHP (W/O safe set)Backbone=Llama3 8B Instruct2025.0466.44
NPO (W/O safe set)Backbone=Llama3 8B Instruct2025.0466.29
WHPBackbone=Llama3 8B Instruct2025.0466.16
NPOBackbone=Llama3 8B Instruct2025.0466.13
TABackbone=Llama3 8B Instruct2025.0466.09
SFTBackbone=Llama3 8B Instruct2025.0466.01
AdaSteerModel=LLAMA3.1-8B-INST2025.1066
OriginalBackbone=Llama3 8B Instruct2025.0465.89
CB*Backbone=Llama3 8B Instruct, publicly_available_safety_tuned=true2025.0465.77
CBBackbone=Llama3 8B Instruct2025.0465.77
DPOBackbone=Llama3 8B Instruct2025.0465.31
REPBENDBackbone=Llama3 8B Instruct2025.0465.08
Safety PromptingBackbone=Llama3 8B Instruct2025.0464.41
SFTBase Model=Llama-3.1-8B-IT, Evaluation Metric Type=Acc2025.1063.4
Self-PlayBase Model=Llama-3.1-8B-IT, Evaluation Metric Type=Acc2025.1062.4
Defender-Only + SFTBase Model=Llama-3.1-8B-IT, Evaluation Metric Type=Acc2025.1062.3
Self-Play + SFTBase Model=Llama-3.1-8B-IT, Evaluation Metric Type=Acc2025.1062.3
Block-R1Source Domain=Multi-domain, Backbone=LLaDA-8B-Instruct2026.0562.22
Defender-OnlyBase Model=Llama-3.1-8B-IT, Evaluation Metric Type=Acc2025.1062.2
system promptModel=LLAMA3.1-8B-INST2025.1062
OriginalBackbone=Mistral 7B Instruct-v0.22025.0460.18
WHP (W/O safe set)Backbone=Mistral 7B Instruct-v0.22025.0460.08
CB*Backbone=Mistral 7B Instruct-v0.2, publicly_available_safety_tuned=true2025.0460.08
CBBackbone=Mistral 7B Instruct-v0.22025.0459.93
NPOBackbone=Mistral 7B Instruct-v0.22025.0459.61
REPBENDBackbone=Mistral 7B Instruct-v0.22025.0459.48
R2D2*Backbone=Mistral 7B Instruct-v0.2, publicly_available_safety_tuned=true2025.0459.44
Safety PromptingBackbone=Mistral 7B Instruct-v0.22025.0459.26
WHPBackbone=Mistral 7B Instruct-v0.22025.0459.24
DPOBackbone=Mistral 7B Instruct-v0.22025.0459.21
NPO (W/O safe set)Backbone=Mistral 7B Instruct-v0.22025.0459.12
TABackbone=Mistral 7B Instruct-v0.22025.0458.57
SFTBackbone=Mistral 7B Instruct-v0.22025.0458.46
RMUBackbone=Llama3 8B Instruct2025.0458.23
REPBENDModel=Gemma2 2B Instruct2025.0458.14
Original Weight ModelModel=Gemma2 2B Instruct2025.0457.98
QWEN 3 1.7BModel=QWEN 3 1.7B2025.1057.9
ELSModel=QWEN 3 1.7B2025.1057.9
AdaSteerModel=QWEN 3 1.7B2025.1057.4
SurgicalModel=QWEN 3 1.7B2025.1057.2
CASTModel=QWEN 3 1.7B2025.1056.8
AlphaSteerModel=QWEN 3 1.7B2025.1056.8
RMUBackbone=Mistral 7B Instruct-v0.22025.0450.02
system promptModel=QWEN 3 1.7B2025.1049.1
LLAMA2-7B-CHATModel=LLAMA2-7B-CHAT2025.1047.6
ELSModel=LLAMA2-7B-CHAT2025.1047.6
system promptModel=LLAMA2-7B-CHAT2025.1047.5
SurgicalModel=LLAMA2-7B-CHAT2025.1047
AdaSteerModel=LLAMA2-7B-CHAT2025.1046
AlphaSteerModel=LLAMA2-7B-CHAT2025.1046
CASTModel=LLAMA2-7B-CHAT2025.1045.6