List your company
ResearchFormerly WizWand

Accuracy on AIME 24 (Reasoning)

86.3Accuracy on AIME 24Qwen3-Base SAT
-0.85221.77444.467.026May 8, 2025Jul 4, 2025Aug 30, 2025Oct 27, 2025Dec 23, 2025Feb 18, 2026Apr 17, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Qwen3-Base SATSize=3×4B2026.0486.3
TF-TTCLModel=DeepSeek-V3.22026.0483.33
Qwen3 (thinking)Size=32B2026.0481.4
Qwen3-A3BSize=30B2026.0480.4
AWQBackbone=Qwen3-14B, Type=linear, Zero-shot=true2025.1180
TF-GRPOModel=DeepSeek-V3.22026.0480
OpenAI o3-mini (medium)Size=/2026.0479.6
QwQSize=32B2026.0479.5
ParoQuantBackbone=Qwen3-14B, Type=linear, Zero-shot=true2025.1177.8
QTIPBackbone=Qwen3-14B, Type=vector, Zero-shot=true2025.1176.7
TF-TTCLModel=Qwen-Plus2026.0476.67
FP16Backbone=Qwen3-4B, Zero-shot=true2025.1175.6
FP16Backbone=Qwen3-8B, Zero-shot=true2025.1175.6
ParoQuantBackbone=Qwen3-8B, Type=linear, Zero-shot=true2025.1175.6
ParoQuantBackbone=Qwen3-4B, Type=linear, Zero-shot=true2025.1173.3
FP16Backbone=Qwen3-14B, Zero-shot=true2025.1173.3
QTIPBackbone=Qwen3-8B, Type=vector, Zero-shot=true2025.1172.2
AWQBackbone=Qwen3-8B, Type=linear, Zero-shot=true2025.1172.2
E-QATBackbone=Qwen3-14B, Type=linear, Zero-shot=true2025.1171.1
Qwen3-Base Debate + JudgeSize=3×8B2026.0471.1
Llama 3.1-Base SATSize=3×8B2026.0470.3
Chain-of-ThoughtModel=DeepSeek-V3.22026.0470
TF-GRPOModel=Qwen-Plus2026.0470
DeepSeek-R1 Distill LlamaSize=70B2026.0470
Qwen3-Base Role-playSize=3×8B2026.0469.9
E-QATBackbone=Qwen3-8B, Type=linear, Zero-shot=true2025.1168.9
Qwen3-Base DebateSize=3×8B2026.0468.7
QTIPBackbone=Qwen3-4B, Type=vector, Zero-shot=true2025.1167.8
Base LLMModel=DeepSeek-V3.22026.0466.67
DS-Qwen 14BModel Size=14B, 0-shot=true2025.0563.33
AWQBackbone=Qwen3-4B, Type=linear, Zero-shot=true2025.1162.2
Layer-wise CapreseModel Size=14B, Sparse Algorithm=CATS, 0-shot=true2025.0561.67
CATSModel Size=14B, 0-shot=true2025.0558.34
E2E CapreseModel Size=14B, Sparse Algorithm=CATS, 0-shot=true2025.0558.34
E-QATBackbone=Qwen3-4B, Type=linear, Zero-shot=true2025.1153.3
DS-Qwen 7BModel Size=7B, 0-shot=true2025.0547.5
FP16Backbone=R1-Distill-Llama-8B, Zero-shot=true2025.1142.2
Layer-wise CapreseModel Size=14B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.0541.67
Qwen3-Base-DAPOSize=8B2026.0441.3
Chain-of-ThoughtModel=Qwen-Plus2026.0440
E2E CapreseModel Size=14B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.0539.17
Qwen3-Base-GRPOSize=8B2026.0439.1
QTIPBackbone=R1-Distill-Llama-8B, Type=vector, Zero-shot=true2025.1137.8
ParoQuantBackbone=R1-Distill-Llama-8B, Type=linear, Zero-shot=true2025.1136.6
Layer-wise CapreseModel Size=7B, Sparse Algorithm=CATS, 0-shot=true2025.0535
AWQBackbone=R1-Distill-Llama-8B, Type=linear, Zero-shot=true2025.1134.4
CATSModel Size=7B, 0-shot=true2025.0534.17
E2E CapreseModel Size=7B, Sparse Algorithm=CATS, 0-shot=true2025.0533.33
GRIFFINModel Size=14B, 0-shot=true2025.0532.5
E-QATBackbone=R1-Distill-Llama-8B, Type=linear, Zero-shot=true2025.1132.2
Qwen3-BaseSize=14B2026.0431.7
Qwen3-BaseSize=32B2026.0431
DS-Qwen 1.5BModel Size=1.5B, 0-shot=true2025.0530
E2E CapreseModel Size=7B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.0530
Base LLMModel=Qwen-Plus2026.0430
Layer-wise CapreseModel Size=7B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.0529.17
GRIFFINModel Size=7B, 0-shot=true2025.0521.67
E2E CapreseModel Size=1.5B, Sparse Algorithm=CATS, 0-shot=true2025.0520.83
Qwen2.5-BaseSize=72B2026.0418.9
Layer-wise CapreseModel Size=1.5B, Sparse Algorithm=CATS, 0-shot=true2025.0515.83
CATSModel Size=1.5B, 0-shot=true2025.0511.67
GPT-4o-mini-2024-07-18Size=/2026.048.1
E2E CapreseModel Size=1.5B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.056.67
GRIFFINModel Size=1.5B, 0-shot=true, Sparsity=50%2025.052.5
Layer-wise CapreseModel Size=1.5B, Sparse Algorithm=GRIFFIN, 0-shot=true2025.052.5