List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 25 (Mean@32)

89.58Mean@32FlashMLA
-2.82421.165545.15569.1445Aug 25, 2025Oct 5, 2025Nov 15, 2025Dec 26, 2025Feb 5, 2026Mar 18, 2026Apr 29, 2026
Updated 6mo ago

Evaluation results

MethodLinks
FlashMLABackbone=LongCat-Flash-thinking, Precision=BF162026.0289.58
SnapMLABackbone=LongCat-Flash-thinking, Precision=FP82026.0288.44
FlashMLABackbone=DeepSeek-V3.1, Precision=BF162026.0287.92
SnapMLABackbone=DeepSeek-V3.1, Precision=FP82026.0285.42
SFTBackbone=Qwen2.5-7B-Instruct2025.0823.02
PSFTwarm-upBackbone=Qwen2.5-7B-Instruct2025.0823.02
PSFTBackbone=Qwen2.5-7B-Instruct2025.0821.98
SFT-KLBackbone=Qwen2.5-7B-Instruct2025.0821.56
PSFTwarm-upBackbone=Llama3.1-8B-Instruct2025.0818.75
SFTBackbone=Llama3.1-8B-Instruct2025.0816.77
SFT-KLBackbone=Llama3.1-8B-Instruct2025.0816.15
GRPO + EntrocraftTraining Samples=400K2026.0415.7
GRPO + Clip-CovTraining Samples=400K2026.0415
GRPO + EntrocraftTraining Samples=300K2026.0414.8
PSFTBackbone=Llama3.1-8B-Instruct2025.0814.48
GRPO + Clip-CovTraining Samples=300K2026.0414.2
GRPO + EntrocraftTraining Samples=200K2026.0414.2
GRPO + Clip-HigherTraining Samples=400K2026.0413.6
GRPO + EntrocraftTraining Samples=100K2026.0413.1
GRPO + Clip-HigherTraining Samples=300K2026.0412.8
GRPO + Clip-CovTraining Samples=200K2026.0412.5
GRPO + Clip-HigherTraining Samples=200K2026.0412.3
GRPO + Clip-CovTraining Samples=100K2026.0410.9
GRPOTraining Samples=200K2026.0410.1
GRPOTraining Samples=100K2026.049.9
GRPOTraining Samples=300K2026.049.6
GRPO + Clip-HigherTraining Samples=100K2026.049.6
GRPOTraining Samples=400K2026.049.4
BaseBackbone=Qwen2.5-7B-Instruct2025.088.75
BaseBackbone=Llama3.1-8B-Instruct2025.080.73