List your company

Question Answering on MedQA USMLE

87.4AccuracyMeerkat
20.21637.65855.172.542Nov 16, 2022Jun 10, 2023Jan 3, 2024Jul 27, 2024Feb 19, 2025Sep 13, 2025Apr 8, 2026

Evaluation results

MethodLinks
MeerkatModel size=70B2024.0387.4—
GPT-4Shots=5-shot2024.0386.6—
MeerkatModel size=8B2024.0374.2—
MeerkatModel size=7B2024.0370.3—
Reflection-ToTTeacher Model=LLaMA3-70B, Student Model=LLaMA3.1-8b, Question Used=1000, Generated Data=1514 (pairs)2025.0466.44—
CoTTeacher Model=QwQ, Student Model=LLaMA3.1-8b, Question Used=10178, Generated Data=6659 (samples)2025.0465.01—
CoT (Matched Pairs)Teacher Model=QwQ, Student Model=LLaMA3.1-8b, Question Used=10178, Generated Data=1514(pairs)2025.0464.73—
OriginalBackbone Model=Lingshu-7B2025.1263.47—
ModelMergeBackbone Model=Lingshu-7B2025.1262.69—
PSIBackbone Model=Lingshu-7B2025.1262.18—
BaselineStudent Model=LLaMA3.1-8b2025.0461.01—
LLaMA-Med42Retrieval=Dense RAG, Reformulation=On, Reranker=On, Runtime(s)=843.72026.0460.4957.8
LLaMA-Med42Retrieval=Dense RAG, Reformulation=On, Reranker=Off, Runtime(s)=761.52026.0460.0257.32
LLaMA-Med42Retrieval=Hybrid RAG, Reformulation=On, Reranker=On, Runtime(s)=3079.22026.0459.3156.61
LLaMA-Med42Retrieval=Dense RAG, Reformulation=Off, Reranker=Off, Runtime(s)=387.52026.0459.0756.37
Gemma3Retrieval=Dense RAG, Reformulation=On, Reranker=On, Runtime(s)=1149.22026.0458.655.9
RESTABackbone Model=Lingshu-7B2025.1258.52—
GPT-3.5Shots=5-shot2024.0358.5—
ModelMergeBackbone Model=HuatuoGPT-Vision-7B2025.1255.62—
OriginalBackbone Model=HuatuoGPT-Vision-7B2025.1253.26—
PSIBackbone Model=HuatuoGPT-Vision-7B2025.1253.1—
BioMistralModel size=7B2024.0351.4—
RESTABackbone Model=HuatuoGPT-Vision-7B2025.1248.7—
MediTronModel size=7B2024.0344.6—
GalacticaModel Size=120B, Shots=0, Domain=out-of-domain2022.1144.4—
BLOOMShots=5, Domain=out-of-domain2022.1123.3—
OPTShots=5, Domain=out-of-domain2022.1122.8—