Atlas

Benchmarks

← All benchmarks

MedMCQA (Open Medical-LLM)

Science · 2024-04-19

MedMCQA collects Indian medical entrance-exam questions across 21 subjects. The Open Medical-LLM Leaderboard scores the 4,183-question validation split under lm-evaluation-harness with a fixed prompt, a different protocol from lab-reported runs, so it forms its own factor group. Four answer options give a 25 percent guessing floor.

Top models (higher is better)

ModelScore
Llama 3 8B57.5
Llama 3 8B Instruct56.8
Qwen1.5 7B50.2
Gemma 7B49.3
Mistral 7B v0.1 Base48.2
Solar 10.7B Instruct V1.047.5
Zephyr 7B Beta45.7
Qwen1.5 7B Chat45.3
Gemma 1.1 7B IT44.4
Gemma 7B IT42.2
Mistral 7B Instruct v0.141.2
Phi-1.534.3
Gemma 2B30.2
Falcon 7B27.2
Loading Atlas data…