Overview of BioASQ 2026: The fourteenth BioASQ Challenge on Large-Scale Biomedical Semantic Indexing and Question Answering

Anastasios Nentidis; Georgios Katsimpras; Anastasia Krithara; Martin Krallinger; Miguel Rodríguez-Ortega; Eduard Rodriguez-López; Natalia Loukachevitch; Igor Rozhkov

cs.CLSep 30, 2026

Overview of BioASQ 2026: The fourteenth BioASQ Challenge on Large-Scale Biomedical Semantic Indexing and Question Answering

Authors: Anastasios Nentidis, Georgios Katsimpras, Anastasia Krithara, Martin Krallinger, Miguel Rodríguez-Ortega, Eduard Rodriguez-López, Natalia Loukachevitch, Igor Rozhkov, +13 more

Organizations: National Center for Scientific Research “Demokritos”, Athens, Greece · Barcelona Supercomputing Center, Barcelona, Spain · Lomonosov Moscow State University, Russia · Artificial Intelligence Research Institute, Russia · HSE University, Russia · Aristotle University of Thessaloniki, Greece · Northwell Health, New Hyde Park, New York, USA · Archimedes, Athena Research Center, Greece · University of Padua, Italy

Abstract

This paper presents an overview of the fourteenth edition of the BioASQ challenge, organized in the context of the Conference and Labs of the Evaluation Forum (CLEF) 2026. BioASQ is an international challenge series that supports progress in biomedical language processing tasks ranging from semantic indexing and information extraction to question answering and summarization. In 2026, BioASQ included six shared tasks: a) Task 14b on biomedical semantic question answering. b) Task Synergy14 on question answering for developing biomedical top- ics. c) Task MultiClinSum-2 on multilingual clinical summarization. d) Task BioNNE-R on extracting relations between nested named entities in Russian and English. e) Task ELCardioCC on clinical coding in cardiology. f) Task GutBrainIE on gut-brain interplay information extrac- tion. Across these six tasks, 87 distinct teams participated, submitting more than 1000 runs overall. As in previous editions, several submissions reached competitive performance, reflecting the continued progress of state-of-the-art methods across biomedical language processing tasks.

Figures & tables

Batch	Size	Yes/No	List	Factoid	Summary	Documents	Snippets
Train	5729	1541	1130	1695	1363	10.74	13.97
Test 1	80	17	21	23	19	5.25	12.41
Test 2	80	21	13	20	26	3.64	7.98
Test 3	60	11	17	17	15	3.40	6.95
Test 4	60	16	20	11	13	3.78	8.35
Total	6009	1606	1201	1766	1436	10.43	13.75

Table 1: Statistics on the training and test datasets of task 14b. The numbers for the documents and snippets refer to averages per question.

		Mean Word Count		Mean Sent. Count
Language	Nr. Pairs	Full case	Summary	Full case	Summary
Catalan (ca)	27699	604.80	118.56	27.49	5.59
Czech (cs)	27488	469.89	86.37	30.78	5.73
Danish (da)	27652	485.14	93.76	28.75	5.70
German (de)	27424	510.20	96.89	30.47	5.82
Greek (el)	26749	547.05	100.84	29.53	5.62

Table 2: Descriptive statistics per language sub-track of the MultiClinSum-2 dataset. For each language, the number of case-summary pairs and mean word and sentence counts are reported for both full clinical cases and their corresponding reference summaries.

Split	Docs	Entities	Rels
[RU]
train	716	37,879	23,773
dev	50	3,210	2,521
test	154	9,339	7,078
[EN]
train	55	3,861	3,193

Table 3: BioNNE-R Shared Task general document statistics. Entities are annotated entity mentions; relations are gold relation annotations.

Collection	# Docs	# Entities	Ents/Doc	# Rels	Rels/Doc
Train Gold	639	20530	32.13	8556	13.39
Train Silver	1310	41409	31.61	21523	16.43
Train Bronze	2972	89987	30.28	29692	9.99
Development Set	80	2521	31.51	1261	15.76
Test Set	80	2850	35.62	1285	16.06

Table 5: Dataset statistics for GutBrainIE .

Team	EN	ES	FR	PT	IT	RU	CA	NO	DA	RO	DE	EL	NL	CS	SV	Total
ixa-sum	5	5	2	2	2	—	5	—	—	2	2	—	2	—	—	27
MediScribes	5	—	—	—	—	—	—	—	—	—	—	—	5	—	—	10
NLP4Health	3	3	3	2	—	—	—	—	—	—	—	—	—	—	—	11
DACHausa	1	1	1	1	1	1	1	—	1	1	1	1	1	1	1	14
InfoLab-FEUP	3	—	—	3	—	—	—	—	—	—	—	—	—	—	—	6
PMH-Aqeel	1	1	—	1	1	1	1	—	—	—	—	—	—	—	1	7

Table 6: Participation overview of MultiClinSum-2 task per team and language subtrack. Numbers indicate runs submitted; — indicates no participation.

Resp. type (Phase)	Quest. type	Official measure
Documents (A)	All	Mean Average Precision (MAP)
Snippets (A)	All	F1 (based on character overlaps)
	List	F1
Exact ans. (A+ & B)	Yesno	macro F1 on “yes” & “no” classes
	Factoid	Mean Reciprocal Rank (MRR)
Ideal ans. (A+ & B)	All	Manual scores for precision, recall, repetition, readability

Table 7: The evaluation measures for task 14b per response type and question type [ 39 ] .

R	Qs	AR	Top MAP	Top F1 Snip.	Top MRR	Top F1 list	Top macro-F1
1	63	0	0.538	0.340	-	-	-
2	66	42	0.337	0.262	0.545	1.000	1.000
3	64	55	0.249	0.390	0.394	0.818	0.771
4	48	47	0.315	0.197	0.500	0.900	0.890

Table 10: The number of questions (Qs) and “Answer Ready” (AR) questions and the top system performance per round (R) in Task Synergy14. Retrieval of documents (Top MAP) and snippets (Top F1). Generation of exact factoid (Top MRR), list (Top F1), and yes/no (Top macro-F1) answers.

Lang.	Team	Run	R-1 (R-2)	R-L	BERT Score	Faith.	Com.	Fl.	Cons.
EN	ixa-sum	2	0.44 (0.22)	0.32	0.88	0.73	0.83	0.70	0.70
	NLP4Health	1	0.38 (0.16)	0.27	0.87	0.76	0.94	0.70	0.70
ES	ixa-sum	2	0.47 (0.23)	0.31	0.88	0.71	0.80	0.70	0.65
	NLP4Health	1	0.41 (0.18)	0.27	0.87	0.72	0.84	0.70	0.63
PT	NLP4Health	1	0.38 (0.16)	0.26	0.87	0.74	0.89	0.70	0.64
	ixa-sum	1	0.38 (0.15)	0.26	0.87	0.74	0.91	0.70	0.65

Table 11: Top-2 participant teams per MultiClinSum-2 sub-track, ranked by BERTScore. Developed baseline system results are not included in the ranking (R-1 = ROUGE-1, R-2 = ROUGE-2, R-L = ROUGE-Lsum, Faith. = Faithfulness, Com. = Completeness, Fl. = Fluency, Cons. = Consistency).

Team	English	Russian	Bilingual
ELiRF-UPV	0.5060	0.4717	0.4540
LODAC-NII	0.5016	0.5283	0.5015
aakobiakova	0.4951	0.4403	0.4716
HSE NLP	0.4733	0.5057	0.4527
rabiaozdemir	0.4620	–	–
savvafq	0.4570	0.4829	0.4849

Table 12: Official BioNNE-R results: macro-F1 of each team’s best submission per subtask. Best result per subtask in bold; – indicates no submission.

Team	System	Recall	Precision	Micro-F1
stanimeros	ensemble_metaheuristic_p4_winner…	0.850979	0.882995	0.866691
Georgios_1	Baymax_submission_v1	0.865845	0.861783	0.863809
LSI_UNED	test_set_NER_greekuncased_greek…	0.818709	0.841595	0.829994
alikibliona	submission_A_reranker	0.818347	0.824626	0.821474
elcardiocc	baseline_bert_base_uncased_greek	0.659173	0.903579	0.762264
Pakakisd	max_micro_curriculum	0.809282	0.720000	0.762035

Table 13: Performance of participating systems in the ELCardioCC 2026

Team ID	Run name	Precision	Recall	F1
TWIX [ 43 ]	merged6	0.9548	0.8058	0.8740
NightSun [ 36 ]	run21	0.8756	0.8110	0.8420
Graphwise [ 59 ]	18	0.8613	0.8169	0.8385
TEXA [ 69 ]	1	0.7980	0.8302	0.8138
unibuc-bionlp-bp	1	0.7816	0.8425	0.8109
SMTE [ 45 ]	R1	0.8109	0.7932	0.8019

Table 14: GutBrainIE performance metrics of each team’s top run beating the baseline for NER. The best result is in bold, the second-best is underlined (micro-averaged).

Team ID	Run name	Precision	Recall	F1
TWIX [ 43 ]	24merged6	0.7527	0.6353	0.6890
NightSun [ 36 ]	run16	0.6290	0.6053	0.6169
Graphwise [ 59 ]	9	0.6217	0.5897	0.6053
TUGW [ 27 ]	EXP4SUB1EL	0.7723	0.4437	0.5636
MindGut link	BiomedBERT	0.5738	0.5330	0.5527
GetGut@AAU [ 26 ]	1	0.5515	0.5519	0.5517

Table 15: GutBrainIE performance metrics of each team’s top run beating the baseline for NERD. The best result is in bold, the second-best is underlined (micro-averaged).

Team ID	Run name	Precision	Recall	F1
TWIX [ 43 ]	mre10	0.8996	0.4651	0.6132
NightSun [ 36 ]	run18	0.4459	0.4593	0.4525
SMTE [ 45 ]	R2	0.4476	0.3997	0.4223
GetGut@AAU [ 26 ]	1	0.4546	0.3658	0.4054
GutHub [ 64 ]	1	0.4347	0.3754	0.4029
Graphwise [ 59 ]	BGPT5515GEXP8	0.3517	0.4497	0.3947

Table 16: GutBrainIE performance metrics of each team’s top run beating the baseline for M-RE. The best result is in bold, the second-best is underlined (micro-averaged).

Team ID	Run name	Precision	Recall	F1
TWIX [ 43 ]	11mre12	0.4903	0.2691	0.3475
NightSun [ 36 ]	run18	0.2484	0.2798	0.2632
Graphwise [ 59 ]	BGPT5515GEXP4	0.2092	0.2963	0.2452
GetGut@AAU [ 26 ]	1	0.2123	0.1926	0.2020
TEXA [ 69 ]	1	0.1556	0.1383	0.1464
BASELINE	Atlop-3stage	0.1403	0.1292	0.1345

Table 17: Performance metrics of each team’s top run beating the baseline for C-RE. The best result is in bold, the second-best is underlined (micro-averaged).

Explore similar work

From Voting to Agent Collaboration: Answer-Type-Aware LLM Pipelines for BioASQ 14b
Jul 7, 2026Taeyun Roh, Eunha Lee, Wonjune Jang +3Biomedical Text Question-Answer Pairs
Department of Computer Science and Engineering, Korea University, Seoul, 02841, Republic of Korea · Department of Mathematics, Myongji University, Yongin, 17058, Republic of Korea · 3AIGEN Sciences, Seoul, 04778, Republic of Korea
Cost-Pragmatic Quality Gating and Selection-Fusion Multi-Model Combiners for BioASQ Phases A+ and B
Jul 15, 2026Dima Galat, Marian-Andrei RizoiuRetrievers Performance Metric
University of Technology Sydney, Australia
Retrieval Augmented Biomedical Question Answering with Weak Question Recovery and Neural Reranking for BioASQ Task 14b
Aug 2, 2026Xueying Zhao, Lee Mai, Balaji AnandganeshCross-Encoder Reranking Text Corpora
Georgia Institute of Technology, North Ave NW, Atlanta, GA 30332

Jul 7, 2026, cs.CLJ/K move · Enter open · S save

From Voting to Agent Collaboration: Answer-Type-Aware LLM Pipelines for BioASQ 14b

Taeyun Roh, Eunha Lee, Wonjune Jang, Sohyun Chung+2

Department of Computer Science and Engineering, Korea University, Seoul, 02841, Republic of Korea · Department of Mathematics, Myongji University, Yongin, 17058, Republic of Korea · 3AIGEN Sciences, Seoul, 04778, Republic of Korea

Biomedical Text Question-Answer Pairs Medical Ontology Large Language Model Pipelines Natural Language Inference Majority Voting

Open paper AskPDF arXiv

Biomedical question answering requires not only accurate extraction of information from scientific literature but also reliable integration of evidence across multiple documents. This study presents a question-type-specific large language model (LLM) framework for BioASQ 14b Task B, designed to improve answer robustness and evidence grounding in biomedical question answering. Rather than applying a single prompting strategy to all questions, the framework selects different inference procedures for yes/no, factoid, and list questions according to their distinct reasoning and evaluation requirements. For yes/no questions, snippet shuffling and self-reflection are used to reduce sensitivity to evidence ordering and improve decision stability. For factoid questions, full-snippet input is combined with chain-of-thought-based in-context learning to support accurate biomedical entity identification. For list questions, a multi-agent architecture is employed, in which evidence extraction, candidate generation, answer verification, and final aggregation are handled collaboratively. Preliminary experiments on BioASQ 13b were used to identify effective inference strategies for each question type, and the resulting framework was subsequently evaluated in the official BioASQ 14b Task B challenge. In the official evaluation, our framework showed competitive performance across multiple batches and achieved first place in the factoid subtask of Batch 4. These results demonstrate the effectiveness of combining question-type-specific inference, ensemble prediction, and agent-based verification for reliable biomedical question answering.

Abstract

Explore similar work

From Voting to Agent Collaboration: Answer-Type-Aware LLM Pipelines for BioASQ 14b

Cost-Pragmatic Quality Gating and Selection-Fusion Multi-Model Combiners for BioASQ Phases A+ and B

Retrieval Augmented Biomedical Question Answering with Weak Question Recovery and Neural Reranking for BioASQ Task 14b