List your company
ResearchFormerly WizWand

Reasoning on OBQA

97.67AccuracyCOSE
8.365231.550154.73577.9199Jan 29, 2026Feb 17, 2026Mar 9, 2026Mar 29, 2026Apr 17, 2026May 7, 2026May 27, 2026
Updated 5mo ago

Evaluation results

MethodLinks
COSEModel=Llama-3.2-3B-Instruct2026.0597.67
BaseModel=Qwen3-4B2026.0597
MAEModel=Qwen3-4B2026.0596.8
COSEModel=Qwen3-4B2026.0596.65
AZRModel=Qwen3-4B2026.0596.6
R-ZeroModel=Qwen3-4B2026.0596.4
MAEModel=Llama-3.2-3B-Instruct2026.0596
COSEModel=Qwen3-0.6B2026.0595.96
R-ZeroModel=Llama-3.2-3B-Instruct2026.0581.39
AZRModel=Llama-3.2-3B-Instruct2026.0580.32
COSEModel=Qwen2.5-3B-Instruct2026.0579.67
MAEModel=Qwen2.5-3B-Instruct2026.0579.39
BaseModel=Llama-3.2-3B-Instruct2026.0579
AZRModel=Qwen2.5-3B-Instruct2026.0577.55
R-ZeroModel=Qwen2.5-3B-Instruct2026.0576.8
MAEModel=Qwen3-0.6B2026.0576.4
R-ZeroModel=Qwen3-0.6B2026.0575.2
AZRModel=Qwen3-0.6B2026.0573.4
BaseModel=Qwen3-0.6B2026.0573
BaseModel=Qwen2.5-3B-Instruct2026.0567.8
SmolLM-1.7B + OursNumber of Parameters=~ 1.7B, Zero-shot=true2026.0431.6
SmolLM-360M + OursNumber of Parameters=~ 360M, Zero-shot=true2026.0430.5
Self-Improving PretrainingTraining Dataset=SlimPajama, Pretraining Objective=Quality2026.0130
LaMini-GPT-124MNumber of Parameters=~ 135M, Zero-shot=true2026.0429.6
Self-Improving PretrainingTraining Dataset=SlimPajama, Pretraining Objective=Factuality2026.0129
SmolLM-135M + OursNumber of Parameters=~ 135M, Zero-shot=true2026.0428.6
Galactica-125MNumber of Parameters=~ 135M, Zero-shot=true2026.0428.2
Self-Improving PretrainingTraining Dataset=RedPajama, Pretraining Objective=Safety2026.0127.4
Llama BaseTraining Dataset=Original Llama Pretraining, Pretraining Objective=Standard next token prediction2026.0127.2
Llama Pretrain BaselineTraining Dataset=SlimPajama, Pretraining Objective=Standard next token prediction2026.0127
Llama Pretrain BaselineTraining Dataset=RedPajama, Pretraining Objective=Safety2026.0126.6
OPT-2.7BNumber of Parameters=~ 1.7B, Zero-shot=true2026.0425
OPT-1.3BNumber of Parameters=~ 1.7B, Zero-shot=true2026.0423.4
Pythia-1.4BNumber of Parameters=~ 1.7B, Zero-shot=true2026.0422.4
Pythia-2.8BNumber of Parameters=~ 1.7B, Zero-shot=true2026.0422
Cerebras-GPT-2.7BNumber of Parameters=~ 1.7B, Zero-shot=true2026.0420.6
Pythia-1BNumber of Parameters=~ 360M, Zero-shot=true2026.0419.6
Pythia-410MNumber of Parameters=~ 360M, Zero-shot=true2026.0417.8
OPT-350MNumber of Parameters=~ 360M, Zero-shot=true2026.0417.6
OPT-125MNumber of Parameters=~ 135M, Zero-shot=true2026.0416.6
Cerebras-GPT-1.3BNumber of Parameters=~ 1.7B, Zero-shot=true2026.0416.6
Pythia-160MNumber of Parameters=~ 135M, Zero-shot=true2026.0416
Cerebras-GPT-256MNumber of Parameters=~ 135M, Zero-shot=true2026.0415.8
Cerebras-GPT-590MNumber of Parameters=~ 360M, Zero-shot=true2026.0415.8
Pythia-70MNumber of Parameters=~ 135M, Zero-shot=true2026.0413.2
Cerebras-GPT-111MNumber of Parameters=~ 135M, Zero-shot=true2026.0411.8