Paper page - AutoMedBench: Towards Medical AutoResearch with Agentic AI Models

Source: https://huggingface.co/papers/2606.01961 Authors:

,

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

AutoMedBench presents a comprehensive benchmark for autonomous medical-AI research that evaluates agent performance across five workflow stages, revealing validation as the weakest stage and highlighting the importance of reliable pipeline execution and verification in medical AI workflows.

Autonomous agentsare increasingly expected to support end-to-endmedical-AI researchworkflows, moving beyond isolated prediction tasks or short-form clinical question answering. However, existing medical agent benchmarks primarily evaluate final outputs, providing limited visibility into agent behavior within the research process. To address this gap, we present AutoMedBench, aworkflow-aware benchmarkfor autonomousmedical-AI researchacross diverse medical imaging and multimodalinferencetasks, organizing agent execution into a unifiedfive-stage workflow(S1-S5):Plan,Setup,Validate,Inference, andSubmit. It compriseslong-horizon taskswith each run averaging 33 agent turns, spanning fiveresearch tracks:segmentation,image enhancement,visual question answering(VQA),report generation, andlesion detection. Each task is evaluated under two difficulty tiers, Lite and Standard, which use the same data and metrics but differ in the amount of task-brief scaffolding, and each run is scored using both final task performance and S1-S5 stage scores, enablingstage-level analysisfrom the initial task brief to the finalsubmitted artifact. Across thousands of recorded runs, stage-level scoring reveals thatValidateis the weakest workflow stage on average, whereasSetupis the strongest, suggesting that current agents are better at making pipelines executable than at verifying their reliability. Post-runerror analysisfurther shows thatverificationandsubmission failuresdominate tagged errors, accounting for 37.7% and 38.1% of fired codes respectively, whereastask-understanding errorsare rare at 0.9%, and runs with one fired error code have a 48% lower overall score than runs with no error code on average.

View arXiv pageView PDFProject pageGitHub32Add to collection

Get this paper in your agent:

hf papers read 2606\.01961

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.01961 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.01961 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.01961 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.