List your company
ResearchFormerly WizWand

Multilingual Reasoning on MT-AIME 24

44.4Accuracy (%)ROSA2
0.51211.90623.334.694Sep 27, 2025Oct 23, 2025Nov 18, 2025Dec 14, 2025Jan 9, 2026Feb 4, 2026Mar 2, 2026
Updated 8mo ago

Evaluation results

MethodLinks
ROSA2Model=Qwen3-8B2026.0344.4
ROSAModel=Qwen3-8B, Update Location=LM Head, Reward Model=Model-based2025.0943.93
ROSAModel=Qwen3-8B, Update Location=Hidden States, Reward Model=Rule-based2025.0942.27
ROSAModel=Qwen3-8B, Update Location=LM Head, Reward Model=Rule-based2025.0940.16
TextGradModel=Qwen3-8B2026.0340
ROSAModel=Qwen3-8B2026.0338.6
ROSA2Model=Qwen2.5-7B-Base2026.0337.8
TextGradModel=Qwen2.5-7B-Base2026.0337.6
ROSAModel=Qwen2.5-7B-Base2026.0337
BaselineModel=Qwen3-8B2025.0930.37
ROSAModel=Qwen2.5-7B-Instruct, Update Location=LM Head, Reward Model=Model-based2025.0930.21
BaselineModel=Qwen3-8B2026.0329.4
ROSAModel=Qwen2.5-7B-Instruct, Update Location=LM Head, Reward Model=Rule-based2025.0927.14
ROSAModel=Qwen2.5-7B-Instruct, Update Location=Hidden States, Reward Model=Rule-based2025.0926.75
ROSA2Model=DeepSeek-R1-Distill-Llama-8B2026.0321.4
BaselineModel=Qwen2.5-7B-Instruct2025.0919.24
ROSAModel=DeepSeek-R1-Distill-Llama-8B2026.0317.2
BaselineModel=Qwen2.5-7B-Base2026.0317
TextGradModel=DeepSeek-R1-Distill-Llama-8B2026.0316.2
ROSA2Model=Qwen3-0.6B-Instruct2026.039.6
ROSAModel=Qwen3-0.6B, Update Location=LM Head, Reward Model=Model-based2025.099.43
ROSAModel=DeepSeek-R1-Distill-Llama-8B, Update Location=LM Head, Reward Model=Model-based2025.098.19
ROSAModel=Qwen3-0.6B, Update Location=Hidden States, Reward Model=Rule-based2025.097.9
ROSAModel=Qwen3-0.6B, Update Location=LM Head, Reward Model=Rule-based2025.097.58
ROSAModel=DeepSeek-R1-Distill-Llama-8B, Update Location=Hidden States, Reward Model=Rule-based2025.097.27
ROSAModel=Qwen3-0.6B-Instruct2026.037.2
TextGradModel=Qwen3-0.6B-Instruct2026.037
ROSAModel=Qwen2.5-0.5B-Instruct, Update Location=LM Head, Reward Model=Model-based2025.096.13
ROSAModel=DeepSeek-R1-Distill-Llama-8B, Update Location=LM Head, Reward Model=Rule-based2025.096.13
ROSAModel=Qwen2.5-0.5B-Instruct, Update Location=Hidden States, Reward Model=Rule-based2025.094.9
BaselineModel=Qwen3-0.6B2025.094.8
BaselineModel=DeepSeek-R1-Distill-Llama-8B2026.034.8
BaselineModel=DeepSeek-R1-Distill-Llama-8B2025.094.73
ROSA2Model=Qwen2.5-0.5B-Instruct2026.034.4
ROSAModel=Qwen2.5-0.5B-Instruct2026.033.8
ROSAModel=Qwen2.5-0.5B-Instruct, Update Location=LM Head, Reward Model=Rule-based2025.093.67
BaselineModel=Qwen2.5-0.5B-Instruct2025.093.48
BaselineModel=Qwen3-0.6B-Instruct2026.033.2
BaselineModel=Qwen2.5-0.5B-Instruct2026.032.6
TextGradModel=Qwen2.5-0.5B-Instruct2026.032.2