List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2025 (avg@32)

68.44Avg@32Qwen3-Next-80B-A3B-Instruct
2.358419.514236.6753.8258Dec 15, 2025Dec 23, 2025Jan 1, 2026Jan 10, 2026Jan 18, 2026Jan 27, 2026Feb 5, 2026
Updated 8mo ago

Evaluation results

MethodLinks
Qwen3-Next-80B-A3B-InstructArchitecture=MoE, # Total Params=80B, # Activated Params=3B2026.0168.44
LongCat-Flash-LiteArchitecture=MoE + NE, # Total Params=68.5B, # Activated Params=2.9B~4.5B2026.0163.23
Kimi-Linear-48B-A3BArchitecture=MoE, # Total Params=48B, # Activated Params=3B2026.0159.58
Gemini 2.5 Flash-Lite2026.0150.1
PMD-MEANBackbone=Qwen3-30B-A3B-Base, tau=0.1, Staleness=162026.0237.19
PMD-MEANBackbone=Qwen3-30B-A3B-Base, tau=0.01, Staleness=162026.0235.1
GRPOBackbone=Qwen3-30B-A3B-Base, tau=null, Staleness=162026.0227.92
PMD-MEANBackbone=Qwen2.5-7B, tau=0.005, Staleness=162026.0219.48
On-policyBackbone=Qwen2.5-7B, tau=null, Staleness=12026.0218.33
PMD-MEANBackbone=Qwen2.5-7B, tau=0.01, Staleness=162026.0217.5
TRAPOTraining Paradigm=Semi-supervised, Labeled Samples Count=4K, Unlabeled Samples Count=12K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1217.1
PMD-MEANBackbone=Qwen2.5-7B, tau=0.02, Staleness=162026.0216.67
Fully SupervisedTraining Paradigm=Supervised, Labeled Samples Count=45K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1215.3
Fully SupervisedTraining Paradigm=Supervised, Labeled Samples Count=4K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1214.8
TRAPOTraining Paradigm=Semi-supervised, Labeled Samples Count=1K, Unlabeled Samples Count=3K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1213.8
Fully SupervisedTraining Paradigm=Supervised, Labeled Samples Count=1K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1213.5
TTRLTraining Paradigm=Unsupervised, Unlabeled Samples Count=45K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1212.7
Token-level EntropyTraining Paradigm=Semi-supervised, Labeled Samples Count=1K, Unlabeled Samples Count=3K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1211.9
Sentence-level EntropyTraining Paradigm=Semi-supervised, Labeled Samples Count=1K, Unlabeled Samples Count=3K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1211.5
Self-certaintyTraining Paradigm=Semi-supervised, Labeled Samples Count=1K, Unlabeled Samples Count=3K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1211.4
Sentence-level EntropyTraining Paradigm=Unsupervised, Unlabeled Samples Count=45K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1210.7
TTRLTraining Paradigm=Semi-supervised, Labeled Samples Count=1K, Unlabeled Samples Count=3K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1210.7
GRPOBackbone=Qwen2.5-7B, tau=null, Staleness=162026.0210.52
Qwen-InstructTraining Paradigm=Original, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1210.2
Self-certaintyTraining Paradigm=Unsupervised, Unlabeled Samples Count=45K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.1210.2
Token-level EntropyTraining Paradigm=Unsupervised, Unlabeled Samples Count=45K, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.129.9
Qwen-BaseTraining Paradigm=Original, Backbone Model=Qwen2.5-Math-7B, Sampling Temperature (T=0.6)=0.62025.124.9