List your company
ResearchFormerly WizWand

Mathematical Reasoning Accuracy on MINERVA

49.3AccuracyGSPO + TAC
8.7419.2729.840.33Oct 29, 2025Dec 4, 2025Jan 9, 2026Feb 15, 2026Mar 23, 2026Apr 28, 2026Jun 4, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GSPO + TACBackbone=DeepSeek-R1-Distill-Qwen-7B, Training steps=30002025.1049.3
GRPO-LEADBackbone=DeepSeek-R1-Distill-Qwen-7B, Training steps=30002025.1047.4
SkyWorkBackbone=DeepSeek-R1-Distill-Qwen-7B, Training steps=30002025.1043.4
DeepMathBackbone=DeepSeek-R1-Distill-Qwen-7B, Training steps=30002025.1042.6
ExGRPOBackbone=DeepSeek-R1-Distill-Qwen-7B, Training steps=30002025.1041
BASELINEBackbone=DeepSeek-R1-Distill-Qwen-7B, Training steps=30002025.1040.4
Eurus-2Backbone=DeepSeek-R1-Distill-Qwen-7B, Training steps=30002025.1038.6
VeriThinkerBackbone=DeepSeek-R1-Distill-Qwen-7B, Training steps=30002025.1034.9
Qwen2.5-1.5B R1-Distilled (base)context size=10242026.0632
Open-RScheckpoints=3, context size=10242026.0631.6
ExGRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Training steps=30002025.1030.9
RREDCoT-Nanocheckpoints=2, context size=10242026.0630.5
GSPO + TACBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Training steps=30002025.1027.2
Open-RS3Backbone=DeepSeek-R1-Distill-Qwen-1.5B, Training steps=30002025.1026.8
BASELINEBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Training steps=30002025.1026.5
DRA-GRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Training steps=30002025.1026.1
STILL-3Backbone=DeepSeek-R1-Distill-Qwen-1.5B, Training steps=30002025.1023.2
DRGRPOBackbone=DeepSeek-R1-Distill-Qwen-7B, Training steps=30002025.1010.3