Evidence Prime took part in CLEF 2025 in Madrid, Spain, from 9 to 12 September. Damian Stachura presented our team’s poster in the BioASQ programme, bringing our work on biomedical question answering to a community focused on evaluating information access systems.
Our BioASQ contribution
The poster, Are Smaller Open-Weight LLMs Closing the Gap to Proprietary Models for Biomedical Question Answering?, was authored by Damian Stachura, Joanna Konieczna and Artur Nowak. It examined how smaller open-weight language models compared with proprietary models in BioASQ Task 13B, Phase B.
The challenge covers several answer formats: yes/no responses, named entities, lists and short summaries. Our experiments explored how to select relevant biomedical text snippets, provide useful examples, constrain answer formats and combine outputs from multiple models. These choices matter because evaluating an answer requires attention to the evidence supplied to the model as well as the form of response the task needs.
Damian also accepted a BioASQ Challenge Award on behalf of the Evidence Prime team. Artur’s conference post shares the award and the poster presented in Madrid.

What this means for evidence synthesis
Biomedical question answering shares practical challenges with evidence review: finding relevant passages, extracting the information a question calls for and assessing whether the resulting answer is useful. CLEF provided an opportunity to discuss those challenges through a shared evaluation task and a published description of our methods.
Our takeaway is that model selection and workflow design need to be evaluated together. A useful assessment should examine the task, supporting context and answer format. The findings in this study are specific to the BioASQ benchmark; performance in a complete screening or data-extraction workflow requires separate validation.
