List your company
ResearchFormerly WizWand

Science Reasoning on GPQA (GPQA Metric)

69.4GPQA ScoreLocal Highest
13.853628.274342.69557.1157Oct 5, 2025Nov 1, 2025Nov 28, 2025Dec 25, 2025Jan 21, 2026Feb 17, 2026Mar 17, 2026
Updated 6mo ago

Evaluation results

MethodLinks
Local HighestStudent Model=Qwen2.5-32B-Instruct, Selection Strategy=Local average log probabilities2025.1069.4
OpenThinker2-32BStudent Model=Qwen2.5-32B-Instruct2025.1064.6
LIMO-32BStudent Model=Qwen2.5-32B-Instruct2025.1062.6
Global HighestStudent Model=Qwen2.5-32B-Instruct, Selection Strategy=Global average log probabilities2025.1061.1
Sky-T1-32B-PreviewStudent Model=Qwen2.5-32B-Instruct2025.1056.6
Original ModelStudent Model=Qwen2.5-32B-Instruct2025.1055.1
Granite-3.3MT=MCS, RL=MC2026.0352.86
Granite-3.3MT=MCS, RL=MCS2026.0351.52
Mistral-7BMT=MCS, RL=MCS2026.0341.75
Mistral-7BMT=MCS, RL=MC2026.0340.91
LLaMA-3.1MT=MCS, RL=MCS2026.0336.03
Granite-3.3MT=MC, RL=MCS2026.0335.69
Granite-3.3MT=MC, RL=MC2026.0335.52
LLaMA-3.1MT=MCS, RL=MC2026.0334.01
Granite-3.3MT=MCS, RL=–2026.0329.12
Mistral-7BMT=MC, RL=MC2026.0329.12
Mistral-7BMT=MC, RL=MCS2026.0328.28
Mistral-7BMT=–, RL=–2026.0326.94
LLaMA-3.1MT=MC, RL=MCS2026.0324.75
Mistral-7BMT=MCS, RL=–2026.0324.07
LLaMA-3.1MT=MC, RL=MC2026.0323.06
Granite-3.3MT=–, RL=–2026.0322.56
LLaMA-3.1MT=MCS, RL=–2026.0321.04
LLaMA-3.1MT=–, RL=–2026.0320.2
LLaMA-3.1MT=MC, RL=–2026.0319.53
Granite-3.3MT=MC, RL=–2026.0319.02
Mistral-7BMT=MC, RL=–2026.0315.99