BioASQ 2026 system uses GPT-5.5 and novel retrieval for medical QA

A research paper details a system designed for the BioASQ Task 14B 2026, focusing on retrieval and multi-model combination strategies. The system employs a hybrid retrieval pipeline incorporating dense models, BM25, and reciprocal rank fusion, alongside an agent-driven approach that decomposes questions for targeted retrieval. This method optimizes re-retrieval costs while improving performance on specific metrics. For answer generation, the system explores selection and fusion techniques for combining multiple language models, with GPT-5.5 showing strong performance on certain metrics. AI

IMPACT This research demonstrates advanced retrieval and multi-model combination techniques that could influence future medical question-answering systems.

RANK_REASON Academic paper detailing a system for a specific task (BioASQ). [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• BioASQ
• BM25
• Europe PubMed Central
• GPT-5.5
• iCite Database Snapshot (Initial)
• PubMed
• reciprocal rank fusion

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →