Position: Open and Closed Large Language Models in Healthcare

TL;DR

This work analyzes the evolving roles of open-source and closed-source LLMs in healthcare, addressing how access to model weights and performance trade-offs shape research and clinical use. It combines two data streams—training activity from the Ecosystem Graph and publication mentions from arXiv, complemented by BERTopic topic modeling—to contrast open versus closed models. The findings reveal exponential growth in open LLMs driven by weight accessibility and fine-tuning, alongside continued dominance of closed LLMs in high-performance healthcare tasks such as radiology and multimodal analysis; open LLMs, however, show strength in mental health and patient communication through domain-specific customization. The study highlights the value of hybrid approaches that combine the strengths of both paradigms for scalable, adaptable, and safe healthcare AI deployment.

Abstract

This position paper analyzes the evolving roles of open-source and closed-source large language models (LLMs) in healthcare, emphasizing their distinct contributions and the scientific community's response to their development. Due to their advanced reasoning capabilities, closed LLMs, such as GPT-4, have dominated high-performance applications, particularly in medical imaging and multimodal diagnostics. Conversely, open LLMs, like Meta's LLaMA, have gained popularity for their adaptability and cost-effectiveness, enabling researchers to fine-tune models for specific domains, such as mental health and patient communication.

Figures (4)

• Figure 1: Growth rates of open and closed LLMs.
• Figure 2: Percentage of arXiv papers mentioning open LLMs or closed LLMs from 2019 onwards, with BERT as a baseline.
• Figure 3: Cumulative ratio (left) and counts (right) of medical papers in LLMs.
• Figure 4: Topic modeling results for medical LLM papers.