List your company
ResearchFormerly WizWand

Correctness Prediction (AUROC) on GSM8K

80.7AUROCQwen 2.5
47.83656.36864.973.432Sep 12, 2025Oct 21, 2025Nov 30, 2025Jan 9, 2026Feb 17, 2026Mar 29, 2026May 8, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Qwen 2.5Classifier=Logistic Regression2026.0580.7
Qwen 2.5Classifier=Gradient Boosting2026.0578.7
DeepSeek R1Classifier=Logistic Regression2026.0578.6
Llama 3.1Classifier=Logistic Regression2026.0578.3
DeepSeek R1Classifier=Gradient Boosting2026.0577.5
Llama 3.2Classifier=Gradient Boosting2026.0576.7
Llama 3.1Classifier=Gradient Boosting2026.0575.8
Llama 3.2Classifier=Logistic Regression2026.0575.8
Qwen 3Classifier=Self-Certainty2026.0572.8
Qwen 3Classifier=Logistic Regression2026.0572.7
DeepSeek R1Classifier=Self-Certainty2026.0570.3
Qwen 2.5Classifier=Self-Certainty2026.0568.9
Qwen 3Classifier=Gradient Boosting2026.0566.5
DirectionModel=Qwen 2.5 7B Instruct, Train dataset=TriviaQA2025.0960.1
Verb. conf.Model=Llama 3.3 70B Instruct2025.0959.8
AssessorModel=Ministral 8B Instruct 2410, Assessor type=logistic regression2025.0959.8
AssessorModel=Qwen 2.5 7B Instruct, Assessor type=logistic regression2025.0958.4
DirectionModel=Mistral 7B Instruct v0.3, Train dataset=TriviaQA2025.0957.9
DirectionModel=Ministral 8B Instruct 2410, Train dataset=TriviaQA2025.0957.8
Verb. conf.Model=Ministral 8B Instruct 24102025.0957.7
AssessorModel=DeepSeek R1 Distill Qwen 32B, Assessor type=logistic regression2025.0957.6
AssessorModel=Llama 3.3 70B Instruct, Assessor type=logistic regression2025.0957.3
Llama 3.2Classifier=Self-Certainty2026.0556.6
AssessorModel=Mistral 7B Instruct v0.3, Assessor type=logistic regression2025.0955.9
AssessorModel=Llama 3.1 8B, Assessor type=logistic regression2025.0955.8
DirectionModel=DeepSeek R1 Distill Qwen 32B, Train dataset=TriviaQA2025.0955.2
Verb. conf.Model=Llama 3.1 8B2025.0954
DirectionModel=Llama 3.1 8B, Train dataset=TriviaQA2025.0953.4
Verb. conf.Model=Mistral 7B Instruct v0.32025.0952.5
Verb. conf.Model=Qwen 2.5 7B Instruct2025.0951.3
Verb. conf.Model=DeepSeek R1 Distill Qwen 32B2025.0950.3
DirectionModel=Llama 3.3 70B Instruct, Train dataset=TriviaQA2025.0949.9
Llama 3.1Classifier=Self-Certainty2026.0549.1