What question did this study set out to answer?

This research aims to evaluate how structured output constraints influence the performance of various large language models (LLMs) in radiology protocol selection.

April 12, 2026European Radiology Experimental0 citationsOpen Access

Influence of structured output constraints on GPT-5-Thinking, Gemini 2.5 Pro, and open-weight LLMs for radiology protocol selection

MBMohammed BahaaeldinUniversity Hospital Bonn SNSebastian NowakUniversity Hospital Bonn OSOleksandra SeidelUniversity Hospital Bonn

Key Points

This research aims to evaluate how structured output constraints influence the performance of various large language models (LLMs) in radiology protocol selection.
Compared performance of GPT-5-Thinking, Gemini 2.5 Pro, and open-weight LLMs.
Evaluated effects of structured prompting on each model's output.
Analyzed protocols selected by proprietary versus open models.
Gemini 2.5 Pro showed improved performance with structured prompting.
GPT-5-Thinking showed decreased performance under similar constraints.
Open-weight models exhibited minimal impact from output constraints, while proprietary models outperformed them in protocol selection.

Abstract

Structured output constraints affect LLM performance differently. Gemini 2.5 Pro benefits from structured prompting, while GPT-5-Thinking declines. Open-weight models show minimal impact from output constraining. Proprietary models outperform open models in radiology protocol selection.

Perguntar à IA

Bookmark

View Full Paper