PFB at EVALITA 2026: Overview of the Prometeia Financial Benchmark

Alessandro Pietro Bardelli, Tolga Çekiç, İrem Demirtaş, Michele Filannino, Simona Scala, Andrea Galassi, Gianmarco Pappacoda, Paolo Torroni

Keywords: Large Language Models, Finance NLP, Multiple-choice QA, Multilingual benchmark, Prometeia Financial Benchmark

TL;DR: The Prometeia Financial Benchmark (PFB) is the EVALITA 2026 shared task on finance questions across 3 languages: Italian, English, and Turkish.

Abstract: The Prometeia Financial Benchmark (PFB) is the EVALITA 2026 shared task on finance questions across 3 languages: Italian, English, and Turkish, and 3 difficulty levels: easy, medium, and hard. The challenge is organized in two subtasks, one on Italian data and one on all three languages. For each subtask, we have received 2 submissions. Our main takeaways are that no significant performance differences stand out across languages and difficulty levels, and that PFB appears to be a challenging benchmark for models smaller than 3B, whereas 20B models already reach an overall accuracy of 90%.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 4