List your company
ResearchFormerly WizWand

Medical Question Answering on MedMCQA

90.4AccuracyProtRLSearch
69.28874.76980.2585.731Nov 16, 2023Apr 17, 2024Sep 17, 2024Feb 17, 2025Jul 20, 2025Dec 20, 2025May 22, 2026
Updated 4mo ago

Evaluation results

MethodLinks
ProtRLSearchReinforcement Learning=true, Protein Pre-training=true2026.0390.4
PADsource_model=Qwen3-32B, training_dataset=gsm8k2026.0289.02
ProtRLSearchProtein Pre-training=false2026.0388.5
PADsource_model=Gemma3-27B-it, training_dataset=gsm8k2026.0288.48
PADsource_model=Qwen3-8B, training_dataset=gsm8k2026.0288.36
DAsource_model=Qwen3-32B, training_dataset=gsm8k2026.0287.49
RSFTsource_model=Qwen3-32B, training_dataset=gsm8k2026.0287.04
RSFTsource_model=Qwen3-8B, training_dataset=gsm8k2026.0286.88
DAsource_model=Qwen3-8B, training_dataset=gsm8k2026.0286.35
RSFTsource_model=Gemma3-27B-it, training_dataset=gsm8k2026.0286.35
DAsource_model=Gemma3-27B-it, training_dataset=gsm8k2026.0286.28
ProtRLSearchReinforcement Learning=false2026.0385.8
o3Size category=Large Models2025.0783.3
Qwen3.5-397BBase Model=Qwen3.5-397B2026.0583.2
o1Model Category=Proprietary2026.0182.7
RandomModel=Llama 8B, Category=Selection2026.0582.4
Kimi-K2.5Base Model=Kimi-K2.52026.0582.3
AcquisitionSynthesis (Format)Model=Llama 8B, Category=AS2026.0581.9
Qwen3.5-122BBase Model=Qwen3.5-122B2026.0581.8
GLM-5.1Base Model=GLM-5.12026.0581.3
Gemini 2.5 ProSize category=Large Models2025.0781.1
AcquisitionSynthesis (Confidence)Model=Llama 8B, Category=AS2026.0581
AcquisitionSynthesis (Gradient Norm)Model=Llama 8B, Category=AS2026.0580.9
Deepseek-V3.2Base Model=Deepseek-V3.22026.0580.7
GPT-4.12025.0780.61
GPT-4.1(2025-04-14)Maximum output tokens=32,7682025.0780.58
AcquisitionSynthesis (Answer Variance)Model=Llama 8B, Category=AS2026.0580.3
Gemini 2.5 FlashSize category=Large Models2025.0779.7
FilteredModel=Llama 8B, Category=Selection2026.0579.3
Qwen3.5-35BBase Model=Qwen3.5-35B2026.0579.2
DeepSeek R1Size category=Large Models2025.0778.8
OriginalModel=Llama 8B, Category=Synthesis2026.0578.8
AcquisitionSynthesis (Diversity)Model=Llama 8B, Category=AS2026.0577.8
AcquisitionSynthesis (Proximity)Model=Llama 8B, Category=AS2026.0577
HuatuoGPT-o1-72BParameter Scale=72B2025.0476.76
SMCSMaximum output tokens=32,7682025.0776.5
CascadeDebateModel Family=Llama-3.22026.0476.33
DataEnvGymModel=Qwen 7B, Category=Synthesis2026.0576.2
GPT-4o(2024-08-06)Maximum output tokens=32,7682025.0776.17
GPT-4oSize category=Large Models2025.0776.1
SEMA-RAGBackbone=kimi-k22026.0576.07
Claude-3.5-Sonnet(2024-06-20)Maximum output tokens=32,7682025.0776
HuatuoGPT-o1-72BMaximum output tokens=32,7682025.0776
Untrained modelModel=Llama 8B2026.0575.8
SMCS2025.0775.69
FilteredModel=Qwen 7B, Category=Selection2026.0575.4
OriginalModel=Qwen 7B, Category=Synthesis2026.0575.3
AcquisitionSynthesis (Answer Variance)Model=Qwen 7B, Category=AS2026.0575.3
HuatuoGPT-o1-70BParameter Scale=70B2025.0475.23
AcquisitionSynthesis (Diversity)Model=Qwen 7B, Category=AS2026.0575.2
SEMA-RAGBackbone=deepseek-v3.1, Retrieval Strategy=multi-round agentic retrieval2026.0575.09
AcquisitionSynthesis (Confidence)Model=Qwen 7B, Category=AS2026.0575
GPT-o3-mini(2025-01-31)Maximum output tokens=32,7682025.0774.92
Self-MoA2025.0774.92
GPT-OSS-120BBase Model=GPT-OSS-120B2026.0574.9
MedAgentsBackbone=GPT-4, Evaluation Protocol=zero-shot setting2023.1174.8
Claude-3.7-Sonnet(2025-02-19)Maximum output tokens=32,7682025.0774.75
AcquisitionSynthesis (Proximity)Model=Qwen 7B, Category=AS2026.0574.5
DRAGBackbone LLM=Phi-3.5-mini-instruct2025.0674.4
OpenBioLLM-70BParameter Scale=70B2025.0474.23
SEMA-RAGBackbone=qwen3-coder-plus2026.0574.23
DRAGBackbone LLM=Llama-3.1-8B-Instruct2025.0674.2
MedGemma 27BSize category=Small Models, test-time scaling=true2025.0774.2
Qwen3.5-9B + MaRBase Model=Qwen3.5-9B, Optimization Method=MaR2026.0574.2
i-MedRAGBackbone=kimi-k22026.0574.13
AcquisitionSynthesis (Format)Model=Qwen 7B, Category=AS2026.0574.1
OpenBioLLM 70BSize category=Large Models2025.0774
MeerkatModel size=70B2024.0373.9
MedCPTBackbone=kimi-k22026.0573.85
Prismatic Syn.Model=Llama 8B, Category=Synthesis2026.0573.8
GPT-4-baseRAG Protocol=No-RAG baselines, Few-shot settings=5-shot2026.0473.7
DRAGBackbone LLM=Llama-3.2-3B-Instruct2025.0673.6
GPT-4oModel Category=Proprietary2026.0173.5
Qwen3.5-9B + DAPOBase Model=Qwen3.5-9B, Optimization Method=DAPO2026.0573.5
Qwen3.5-9BBase Model=Qwen3.5-9B2026.0573.4
MedRAGBackbone=kimi-k22026.0573.2
DeepSeek-R1-Distill-Llama-70BMaximum output tokens=32,7682025.0773.17
Few-shot CoT + SCBackbone=GPT-4, Evaluation Protocol=few-shot setting2023.1173.1
CoTBackbone=qwen3-coder-plus2026.0573.06
UltraMedical-70B-3Parameter Scale=70B2025.0472.94
DRAGBackbone LLM=Qwen2.5-3B-Instruct2025.0672.8
GPT-4Shots=5-shot2024.0372.4
DRAGBackbone LLM=LLaMA-2-7B2025.0672.4
DRAGBackbone LLM=Gemma-2-2B-it2025.0672.4
Med-PaLM 2RAG Protocol=No-RAG baselines2026.0472.3
Llama-3.3-70B-InstructMaximum output tokens=32,7682025.0772.25
CoTBackbone=kimi-k22026.0572.08
CoTBackbone=deepseek-v3.1, Retrieval Strategy=no retrieval2026.0571.69
SEMA-RAGBackbone=gemini-2.0-flash2026.0571.6
PulseMind-72BModel Category=Open-source, Model Scale=~72B2026.0171.3
GLM-Z1-32B-0414Maximum output tokens=32,7682025.0771.08
Qwen3-32BMaximum output tokens=32,7682025.0770.92
Llama-3.3-Nemotron-Super-49B-v1Maximum output tokens=32,7682025.0770.92
QwQ-32BMaximum output tokens=32,7682025.0770.75
MA-PoPAgent Composition (Heterogeneous)=Qwen-32B, Falcon-34B, Gemma-27B, Multi-Agent Strategy=Multi-Agent2026.0570.67
AcquisitionSynthesis (Gradient Norm)Model=Qwen 7B, Category=AS2026.0570.5
i-MedRAGBackbone=qwen3-coder-plus2026.0570.26
PADsource_model=gemma3-27b-it, train_dataset=gsm8k, base_model=Qwen3-1.7B2026.0270.2
Few-shotBackbone=GPT-4, Evaluation Protocol=few-shot setting2023.1170.1
Zero-shot CoT + SCBackbone=GPT-4, Evaluation Protocol=zero-shot setting2023.1170.1