List your company

Mathematical Reasoning on AIME 24 (Avg@32 Accuracy)

93.85Avg@32 AccuracyFlashMLA
17.87837.601557.32577.0485Feb 2, 2026Feb 3, 2026Feb 5, 2026Feb 6, 2026Feb 8, 2026Feb 9, 2026Feb 11, 2026

Evaluation results

MethodLinks
FlashMLABackbone=DeepSeek-V3.1, Precision=BF162026.0293.85
SnapMLABackbone=DeepSeek-V3.1, Precision=FP82026.0293.65
FlashMLABackbone=LongCat-Flash-thinking, Precision=BF162026.0291.87
SnapMLABackbone=LongCat-Flash-thinking, Precision=FP82026.0291.67
PCLBackbone=DeepSeek-R1-Distill-7B, Strategy=PCL, Runtime=50h2026.0252.8
GPSBackbone=DeepSeek-R1-Distill-7B, Strategy=GPS, Runtime=49h2026.0251.8
Dynamic Sampling (Oracle)Backbone=DeepSeek-R1-Distill-7B, Strategy=DS, Runtime=77h2026.0251
Uniform SamplingBackbone=DeepSeek-R1-Distill-7B, Strategy=Uniform, Runtime=40h2026.0250.9
MoPPSBackbone=DeepSeek-R1-Distill-7B, Strategy=MoPPS, Runtime=42h2026.0250.8
GRESOBackbone=DeepSeek-R1-Distill-7B, Strategy=GRESO, Runtime=53h2026.0249.8
DeepSeek-R1-Distill-7BBackbone=DeepSeek-R1-Distill-7B, Strategy=Base2026.0238.8
OTBBackbone=Qwen3-8b-Base, Reasoning Paradigm=Single-Turn2026.0237.29
OPOBackbone=Qwen3-8b-Base, Reasoning Paradigm=Single-Turn2026.0235.31
Dynamic Sampling (Oracle)Backbone=DeepSeek-R1-Distill-1.5B, Strategy=DS, Runtime=30h2026.0234.7
PCLBackbone=DeepSeek-R1-Distill-1.5B, Strategy=PCL, Runtime=17h2026.0234.2
MoPPSBackbone=DeepSeek-R1-Distill-1.5B, Strategy=MoPPS, Runtime=17h2026.0234
GPSBackbone=DeepSeek-R1-Distill-1.5B, Strategy=GPS, Runtime=16h2026.0233.8
OGBBackbone=Qwen3-8b-Base, Reasoning Paradigm=Single-Turn2026.0233.13
Uniform SamplingBackbone=DeepSeek-R1-Distill-1.5B, Strategy=Uniform, Runtime=16h2026.0232.4
GRESOBackbone=DeepSeek-R1-Distill-1.5B, Strategy=GRESO, Runtime=27h2026.0231.5
GRPOBackbone=Qwen3-8b-Base, Reasoning Paradigm=Single-Turn2026.0231.35
RLOOBackbone=Qwen3-8b-Base, Reasoning Paradigm=Single-Turn2026.0228.96
DeepSeek-R1-Distill-1.5BBackbone=DeepSeek-R1-Distill-1.5B, Strategy=Base2026.0220.8