List your company
ResearchFormerly WizWand

Language Understanding on MMLU (Score)

88.6MMLU Scoregpt-oss-120b
13.40832.92952.4571.971Jun 6, 2024Sep 20, 2024Jan 4, 2025Apr 20, 2025Aug 4, 2025Nov 18, 2025Mar 4, 2026
Updated 6mo ago

Evaluation results

MethodLinks
gpt-oss-120bNumber of Parameters=117B, Optimization Level=high2026.0188.6
Solar OpenNumber of Parameters=102B2026.0188.2
gpt-oss-120bNumber of Parameters=117B, Optimization Level=medium2026.0187.9
GLM-4.5-AirNumber of Parameters=110B2026.0183.3
Qwen2-72B-InstructType=Instruction-tuned2024.0782.3
Llama-3-70B-InstructType=Instruction-tuned2024.0782
DenseModel=Qwen-3 14B, Sparsity=02025.1080.47
DenseModel=Qwen-3 8B, Sparsity=02025.1076.82
Qwen1.5-110B-ChatType=Instruction-tuned2024.0776.5
Qwen1.5-72B-ChatType=Instruction-tuned2024.0775.6
Mixtral-8x22B-InstructType=Instruction-tuned2024.0774
ARMORModel=Qwen-3 14B, Sparsity=2:4+3.89%2025.1071.43
SGDPOBase model=Qwen-2 instruct 7B2025.0570.69
SamPOBase model=Qwen-2 instruct 7B2025.0570.67
DPOBase model=Qwen-2 instruct 7B2025.0570.65
BCOBase model=Qwen-2 instruct 7B2025.0570.65
SFTBase model=Qwen-2 instruct 7B2025.0570.6
NCABase model=Qwen-2 instruct 7B2025.0570.57
TDPOBase model=Qwen-2 instruct 7B2025.0570.55
IPOBase model=Qwen-2 instruct 7B2025.0570.39
System PromptBase Model=Llama-3.1-8B2026.0367
ARMORModel=Qwen-3 8B, Sparsity=2:4+5.03%2025.1066.22
Llama-3.1-8B-SFTBase Model=Llama-3.1-8B2026.0366
VCLBase Model=Llama-3.1-8B2026.0365
SparseGPTModel=Qwen-3 14B, Sparsity=2:42025.1064.73
DROBase Model=Llama-3.1-8B2026.0364
Self-CDBase Model=Llama-3.1-8B2026.0363
WandaModel=Qwen-3 14B, Sparsity=2:42025.1062.93
Vector AblationBase Model=Llama-3.1-8B2026.0362
NoWag-PModel=Qwen-3 14B, Sparsity=2:42025.1061.69
NPO+KLBase Model=Llama 3.1 8B, Variant=12025.0561.4
Unilogit+KLBase Model=Llama 3.1 8B, Variant=22025.0561.4
BaselineBase Model=Llama 3.1 8B2025.0560.8
ME+GDBase Model=Llama 3.1 8B2025.0560.8
NPO+KLBase Model=Llama 3.1 8B, Variant=22025.0560.4
NPOBase Model=Llama 3.1 8B2025.0560.2
UnDIAL+KLBase Model=Llama 3.1 8B, Variant=12025.0560.2
GABase Model=Llama 3.1 8B2025.0559.6
SimNPO+KLBase Model=Llama 3.1 8B, Variant=22025.0559.6
RKLD+KLBase Model=Llama 3.1 8B2025.0559.5
SimNPO+KLBase Model=Llama 3.1 8B, Variant=12025.0559.1
GA+KLBase Model=Llama 3.1 8B2025.0558.7
UnDIAL+KLBase Model=Llama 3.1 8B, Variant=22025.0558.5
SparseGPTModel=Qwen-3 8B, Sparsity=2:42025.1055.77
WandaModel=Qwen-3 8B, Sparsity=2:42025.1055.75
VanillaTarget Model=LLaMA2-13B-Chat2024.0654.9
Unilogit+KLBase Model=Llama 3.1 8B, Variant=12025.0554.4
NoWag-PModel=Qwen-3 8B, Sparsity=2:42025.1054.1
PaCETarget Model=LLaMA2-13B-Chat2024.0653.1
Qwen2-1.5Bnumber of parameters=1.5B2024.0752.4
PromptingTarget Model=LLaMA2-13B-Chat2024.0652.4
OrthoProjTarget Model=LLaMA2-13B-Chat2024.0651.1
VecAddTarget Model=LLaMA2-13B-Chat2024.0650.9
Qwen1.5-1.8Bnumber of parameters=1.8B2024.0743.7
VanillaTarget Model=LLaMA2-7B-Chat2024.0643.4
PaCETarget Model=LLaMA2-7B-Chat2024.0638.4
Qwen2-0.5Bnumber of parameters=0.5B2024.0737.9
Qwen1.5-0.5Bnumber of parameters=0.5B2024.0735
OrthoProjTarget Model=LLaMA2-7B-Chat2024.0634.1
MoEModel Size=15B-A1.5B, Training Strategy=Baseline2026.0233.24
Expert Divergence LearningModel Size=15B-A1.5B, Scheme=49-class2026.0233.21
Expert Divergence LearningModel Size=15B-A1.5B, Scheme=3-class2026.0232.98
MoEModel Size=8B-A0.8B, Training Strategy=Baseline2026.0232.54
Expert Divergence LearningModel Size=8B-A0.8B, Scheme=49-class2026.0232.05
Expert Divergence LearningModel Size=8B-A0.8B, Scheme=3-class2026.0231.69
Expert Divergence LearningModel Size=3B-A0.3B, Scheme=3-class2026.0231.47
Expert Divergence LearningModel Size=3B-A0.3B, Scheme=49-class2026.0231.41
MoEModel Size=3B-A0.3B, Training Strategy=Baseline2026.0231.16
VecAddTarget Model=LLaMA2-7B-Chat2024.0630.6
PromptingTarget Model=LLaMA2-7B-Chat2024.0616.3