List your company
ResearchFormerly WizWand

Mathematical Reasoning on Gaokao 2023

76.6AccuracyAgentic-Verifier-Qwen3-4B
14.230.446.662.8Jun 9, 2025Jul 31, 2025Sep 22, 2025Nov 14, 2025Jan 6, 2026Feb 28, 2026Apr 22, 2026
Updated 6mo ago

Evaluation results

MethodLinks
Agentic-Verifier-Qwen3-4BRefinement Turn=22026.0476.6———3.42.3
Agentic-Verifier-Qwen3-4BRefinement Turn=32026.0476.4———2.62.9
Agentic-Verifier-Qwen3-4BRefinement Turn=12026.0475.6———40.31.8
DS-Distill-14BRefinement Turn=32026.0473———4.41.6
DS-Distill-14BRefinement Turn=22026.0470.1———4.71.6
Qwen3-4B-ThinkRefinement Turn=32026.0469.6———2.31.6
Qwen3-4B-ThinkRefinement Turn=22026.0469.1———5.51.3
AbstRaLModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=AbstRaL, Evaluation Protocol=Zero-shot2025.0668.3—————
Ori-SFTModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=Ori-SFT, Evaluation Protocol=Zero-shot2025.0667.5—————
CoT-RLModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=CoT-RL, Evaluation Protocol=Zero-shot2025.0667.5—————
CoAModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=CoA, Evaluation Protocol=Zero-shot2025.0667—————
DS-Distill-14BRefinement Turn=12026.0467———32.52.6
Qwen3-4B-ThinkRefinement Turn=12026.0464.9———31.23.4
HiPO-Mt-biasBackbone=Qwen2.5-7B-Instruct2026.0462.18—————
HiPO-Rq+Mt-biasBackbone=Qwen2.5-7B-Instruct2026.0461.11—————
HiPO-Rq-biasBackbone=Qwen2.5-7B-Instruct2026.0459.49—————
BaseBackbone=Qwen2.5-7B-Instruct2026.0458.61—————
DPOBackbone=Qwen2.5-7B-Instruct2026.0457.53—————
Mistral-Small-24B-InstructRefinement Turn=22026.0452.7———4.91.3
Qwen2.5-7B-InstructRefinement Turn=32026.0452.2———3.641.3
Mistral-Small-24B-InstructRefinement Turn=32026.0452.2———2.082.6
Qwen2.5-7B-InstructRefinement Turn=22026.0449.9———5.22.3
HiPO-Rq-biasBackbone=Llama-3.1-8B-Instruct2026.0449.19—————
Mistral-Small-24B-InstructRefinement Turn=12026.0449.1———15.63.6
Llama3.1-8B-InstructRefinement Turn=22026.0448.1———5.22.1
HiPO-Mt-biasBackbone=Llama-3.1-8B-Instruct2026.0447.96—————
Qwen2.5-7B-InstructRefinement Turn=12026.0447———12.72.9
Llama3.1-8B-InstructRefinement Turn=32026.0447———1.32.3
HiPO-Rq+Mt-biasBackbone=Llama-3.1-8B-Instruct2026.0446.44—————
BaseBackbone=Llama-3.1-8B-Instruct2026.0446.03—————
Llama3.1-8B-InstructRefinement Turn=12026.0444.9———10.72.9
DPOBackbone=Llama-3.1-8B-Instruct2026.0444.73—————
AbstRaLModel=Qwen2.5-0.5B-Instruct, GSM Train Method=AbstRaL, Evaluation Protocol=Zero-shot2025.0629.1—————
Ori-SFTModel=Qwen2.5-0.5B-Instruct, GSM Train Method=Ori-SFT, Evaluation Protocol=Zero-shot2025.0627—————
CoAModel=Qwen2.5-0.5B-Instruct, GSM Train Method=CoA, Evaluation Protocol=Zero-shot2025.0623.9—————
CoT-RLModel=Qwen2.5-0.5B-Instruct, GSM Train Method=CoT-RL, Evaluation Protocol=Zero-shot2025.0616.6—————
BiRMBase Model=Qwen2.5-3B2025.03—36.937.838.7——
BiRMBase Model=Qwen2.5-7B2025.03—47.749.150.4——
BiRMBase Model=Llama3.1-8B2025.03—29.43029.6——
ER-PRMBase Model=Qwen2.5-3B2025.03—36.236.335.8——
ER-PRMBase Model=Qwen2.5-7B2025.03—4747.247.3——
ER-PRMBase Model=Llama3.1-8B2025.03—25.726.124.9——
GreedyBase Model=Qwen2.5-3B2025.03—272727——
GreedyBase Model=Qwen2.5-7B2025.03—33.533.533.5——
GreedyBase Model=Llama3.1-8B2025.03—17.117.117.1——
Majority VoteBase Model=Qwen2.5-3B2025.03—35.836.336.1——
Majority VoteBase Model=Qwen2.5-7B2025.03—45.545.445.2——
Majority VoteBase Model=Llama3.1-8B2025.03—26.527.227.1——
Math-ShepherdBase Model=Qwen2.5-3B2025.03—36.636.436.1——
Math-ShepherdBase Model=Qwen2.5-7B2025.03—46.44746.5——
Math-ShepherdBase Model=Llama3.1-8B2025.03—25.825.826.2——
ORMBase Model=Qwen2.5-3B2025.03—37.23735.8——
ORMBase Model=Qwen2.5-7B2025.03—43.643.541.3——
ORMBase Model=Llama3.1-8B2025.03—25.425.224.9——
PRMBase Model=Qwen2.5-3B2025.03—37.337.137.2——
PRMBase Model=Qwen2.5-7B2025.03—45.846.247.3——
PRMBase Model=Llama3.1-8B2025.03—28.227.727.3——