List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2024 (Accuracy %)

83.33AIME 2024 AccuracyOriginal
-3.333219.165941.66564.1641May 16, 2026May 20, 2026May 25, 2026May 29, 2026Jun 3, 2026Jun 7, 2026Jun 12, 2026
Updated 4mo ago

Evaluation results

MethodLinks
OriginalModel=Seed-OSS-36B-Instruct, MLP Sparsity=0%2026.0683.33
SpenseGPT+Model=Seed-OSS-36B-Instruct, MLP Sparsity=25%2026.0680
SpenseGPTModel=Seed-OSS-36B-Instruct, MLP Sparsity=37.5%2026.0678.89
SpenseGPT+Model=Qwen3-32B, MLP Sparsity=37.5%2026.0677.78
SpenseGPTModel=Seed-OSS-36B-Instruct, MLP Sparsity=25%2026.0677.78
SpenseGPT+Model=Qwen3-32B, MLP Sparsity=25%2026.0676.67
SpenseGPT+Model=Seed-OSS-36B-Instruct, MLP Sparsity=37.5%2026.0676.67
OriginalModel=Qwen3-32B, MLP Sparsity=0%2026.0675.56
SparseGPTModel=Seed-OSS-36B-Instruct, MLP Sparsity=50%2026.0675.56
SpenseGPTModel=Qwen3-32B, MLP Sparsity=37.5%2026.0673.33
SpenseGPTModel=Qwen3-32B, MLP Sparsity=25%2026.0671.11
SparseGPTModel=Qwen3-32B, MLP Sparsity=50%2026.0668.89
Text-SerializationBackbone=Qwen3-14B, Workers=32026.0663.33
Parallel-SynthesisBackbone=Qwen3-14B, Workers=3, Merging=Weighted averaging (lambda=0.5)2026.0663.33
BaseModel backbone=Qwen3-4B-It, Fine-tuning method=Base2026.0562.9
MCPO-DAPOBackbone=Qwen3-8B2026.0559.86
MCPO-GRPOBackbone=Qwen3-8B2026.0557.71
MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.0556.9
MCPO-DAPOBackbone=Qwen3-4B2026.0556.35
MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.0555
MGSBackbone=Qwen3-8B2026.0554.17
MCPO-GRPOBackbone=Qwen3-4B2026.0554.15
MT-GRPOBackbone=Qwen3-8B2026.0554.05
DAPOBackbone=Qwen3-4B2026.0553.56
DAPOBackbone=Qwen3-8B2026.0553.46
MT-GRPOBackbone=Qwen3-4B2026.0552.7
GRPOBackbone=Qwen3-8B2026.0551.88
CLIPOBackbone=Qwen3-8B2026.0551.56
MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.0551.5
MGSBackbone=Qwen3-4B2026.0551.25
KVLINKBackbone=Qwen3-14B, Workers=32026.0650
CLIPOBackbone=Qwen3-4B2026.0548.33
GRPOBackbone=Qwen3-4B2026.0548.02
MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=02026.0544.4
Base ModelBackbone=Qwen3-8B2026.0539.89
Base ModelBackbone=Qwen3-4B2026.0536.04
MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.0532.1
VotingBackbone=Qwen3-14B, Workers=32026.0630
MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.0529
BaseModel backbone=Qwen3-8B, Fine-tuning method=Base2026.0526
MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.0524.2
SingleBackbone=Qwen3-14B, Workers=12026.0623.33
MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=02026.0517.5
OPSDModel backbone=Qwen3-4B-It, Fine-tuning method=OPSD2026.0515
HyperPruneModel=Qwen3-32B, MLP Sparsity=50%2026.0613.33
CacheBlendBackbone=Qwen3-14B, Workers=32026.0613.33
SFTModel backbone=Qwen3-8B, Fine-tuning method=SFT2026.0512.3
BaseModel backbone=Qwen3-1.7B, Fine-tuning method=Base2026.0511
OPSDModel backbone=Qwen3-8B, Fine-tuning method=OPSD2026.059.4
SFTModel backbone=Qwen3-4B-It, Fine-tuning method=SFT2026.056.7
WandaModel=Qwen3-32B, MLP Sparsity=50%2026.066.67
APEBackbone=Qwen3-14B, Workers=32026.066.67
MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.054.8
MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.052.3
MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.051.9
HyperPruneModel=Seed-OSS-36B-Instruct, MLP Sparsity=50%2026.061.11
SFTModel backbone=Qwen3-1.7B, Fine-tuning method=SFT2026.050.4
OPSDModel backbone=Qwen3-1.7B, Fine-tuning method=OPSD2026.050
MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=02026.050
WandaModel=Seed-OSS-36B-Instruct, MLP Sparsity=50%2026.060