List your company

Science Question Answering on GPQA (Accuracy, Time, Peak)

47.98AccuracyHybridThinker
11.21620.760530.30539.8495Jun 2, 2026

Evaluation results

MethodLinks
HybridThinkerBackbone=Qwen2.5-7B Series2026.0647.9814.842,124
HybridThinkerBackbone=Llama3.1-8B Series2026.0637.8814.551,971
VanillaBackbone=Qwen2.5-7B Series2026.0637.3724.57,184
LightThinkerBackbone=Qwen2.5-7B Series2026.0637.3713.331,969
VanillaBackbone=Llama3.1-8B Series2026.0637.3724.726,824
LightThinkerBackbone=Llama3.1-8B Series2026.0637.3715.591,933
Distill-R1Backbone=Llama3.1-8B Series2026.0630.8118.165,640
CoTBackbone=Qwen2.5-7B Series2026.0628.281.18998
CoTBackbone=Llama3.1-8B Series2026.0624.754.582,230
Vanilla + H2OBackbone=Llama3.1-8B Series2026.0623.2353.471,024
Distill-R1Backbone=Qwen2.5-7B Series2026.0622.2211.894,383
Vanilla + H2OBackbone=Qwen2.5-7B Series2026.0620.247.981,024
Vanilla + SepLLMBackbone=Llama3.1-8B Series2026.0615.6623.691,024
Vanilla + SepLLMBackbone=Qwen2.5-7B Series2026.0612.6340.891,024