List your company
ResearchFormerly WizWand

Scientific Reasoning on GPQA (Mean@1)

68.3Mean@1Agentic Proposing
21.70833.80445.957.996Jan 29, 2026Jan 30, 2026Jan 31, 2026Feb 1, 2026Feb 2, 2026Feb 3, 2026
Updated 8mo ago

Evaluation results

MethodLinks
Agentic ProposingTraining Data=Agentic-Proposer-4B Generated Problems, Training Budget=10,000 trajectories, Optimizer=GRPO2026.0268.3
Qwen3-4B-Instruct-2507Protocol=zero-shot2026.0262
DAREBackbone=Qwen3-1.7B, Method Category=Test-time scaling methods2026.0132.7
TTRLBackbone=Qwen3-1.7B, Method Category=Test-time scaling methods2026.0131.2
GRPOBackbone=Qwen3-1.7B, Method Category=RL methods on training dataset2026.0131.1
RLPRBackbone=Qwen3-1.7B, Method Category=Test-time scaling methods2026.0131
REINFORCE++Backbone=Qwen3-1.7B, Method Category=RL methods on training dataset2026.0130.8
CO-REWARDING-IBackbone=Qwen3-1.7B, Method Category=Test-time scaling methods2026.0130.5
REINFORCEBackbone=Qwen3-1.7B, Method Category=RL methods on training dataset2026.0130.2
CoTBackbone=Qwen2.5-Math-1.5B, Method Category=Prompt methods (training-free)2026.0128.6
DAREBackbone=Qwen2.5-Math-1.5B, Method Category=Test-time scaling methods2026.0128.5
INTUITORBackbone=Qwen3-1.7B, Method Category=Test-time scaling methods2026.0128.5
Raw modelBackbone=Qwen2.5-Math-1.5B, Method Category=Prompt methods (training-free)2026.0127.8
CoTBackbone=Qwen3-1.7B, Method Category=Prompt methods (training-free)2026.0127.6
Raw modelBackbone=Qwen3-1.7B, Method Category=Prompt methods (training-free)2026.0126.8
RLPRBackbone=Qwen2.5-Math-1.5B, Method Category=Test-time scaling methods2026.0126.5
GRPOBackbone=Qwen2.5-Math-1.5B, Method Category=RL methods on training dataset2026.0126.1
TTRLBackbone=Qwen2.5-Math-1.5B, Method Category=Test-time scaling methods2026.0126.1
REINFORCE++Backbone=Qwen2.5-Math-1.5B, Method Category=RL methods on training dataset2026.0125.9
CO-REWARDING-IBackbone=Qwen2.5-Math-1.5B, Method Category=Test-time scaling methods2026.0125.8
REINFORCEBackbone=Qwen2.5-Math-1.5B, Method Category=RL methods on training dataset2026.0125.2
INTUITORBackbone=Qwen2.5-Math-1.5B, Method Category=Test-time scaling methods2026.0123.5