List your company
ResearchFormerly WizWand

Reasoning on Combined Reasoning Benchmarks

59.29Overall AccuracyOpenSIR
34.714841.094947.47553.8551Nov 1, 2025Nov 24, 2025Dec 18, 2025Jan 10, 2026Feb 3, 2026Feb 26, 2026Mar 22, 2026
Updated 4mo ago

Evaluation results

MethodLinks
OpenSIRBackbone=Qwen3-8B, Training Method=OpenSIR2025.1159.29
GRPOmathBackbone=Qwen3-8B, Training Method=GRPOmath2025.1156.18
BaseBackbone=Qwen3-8B, Training Method=Base2025.1155.89
GRPOgsm8kBackbone=Qwen3-8B, Training Method=GRPOgsm8k2025.1155.15
Absolute ZeroBackbone=Qwen3-8B, Training Method=Absolute Zero2025.1154.53
R-ZeroBackbone=Qwen3-8B, Training Method=R-Zero2025.1153.96
WISTBackbone=Qwen3-14B-Base2026.0349.4
SPICEBackbone=Qwen3-14B-Base2026.0348.6
R-ZeroBackbone=Qwen3-14B-Base2026.0347.5
Base ModelBackbone=Qwen3-14B-Base2026.0346.2
OpenSIRBackbone=DeepSeek-R1-Distill-Llama-8B, Training Method=OpenSIR2025.1140.56
GRPOmathBackbone=DeepSeek-R1-Distill-Llama-8B, Training Method=GRPOmath2025.1137.27
GRPOgsm8kBackbone=DeepSeek-R1-Distill-Llama-8B, Training Method=GRPOgsm8k2025.1136.97
BaseBackbone=DeepSeek-R1-Distill-Llama-8B, Training Method=Base2025.1136.31
R-ZeroBackbone=DeepSeek-R1-Distill-Llama-8B, Training Method=R-Zero2025.1136.25
Absolute ZeroBackbone=DeepSeek-R1-Distill-Llama-8B, Training Method=Absolute Zero2025.1135.66