0

Med Agent Bench

A realistic virtual EHR environment to benchmark medical LLM agents on clinical tasks.

Domain
rl-env
License
unknown
Published
Aug 2025

Cite

Notes

Only stored in your browser.

Attribution

Leaderboard scores
prime-hub
Attribution policy →

Top score 84.0% by Qwen3 30B A3B - 2 models reporting

Score history

2
35%51%68%84%100%Apr 25May 25Jun 25Jul 25Qwen3 30B A3B

Top models

2
Med Agent BenchBar chart with 2 bars. Highest value: Qwen3 30B A3B at 84.
2 models

Related tools

1
View all

Implementations, trainers, datasets and scaffolds linked to this eval.

FAQ

What is Med Agent Bench?
A realistic virtual EHR environment to benchmark medical LLM agents on clinical tasks.
What is the current top score on Med Agent Bench?
The top reported score is 84.0% by Qwen3 30B A3B, across 2 models reporting.
How can a model improve its Med Agent Bench score?
Tools linked to Med Agent Bench on Sophon include Agent Bench RL Env (Prime Intellect) - RL environments, datasets, and scaffolds that target this eval.
What license is Med Agent Bench under?
Med Agent Bench is available under unknown.