📢  18 March 2026  Results announced

📅  15 March 2026  —  End of the test phase 

📅  10 March 2026  —  Beginning of the test phase

📅  21 December 2025 —  Dataset released and available for download

       📂  To access the dataset, register here. 

       ⬇️  Download data 

📅 15 December 2025  — The official QIAS 2026 website is now live! 

📅 18 November 2025   — The QIAS 2026 proposal has been accepted! 

QIAS 2026 is the new edition of the QIAS Shared Task series, dedicated to evaluating Large Language Models (LLMs) in the domain of Islamic studies. This edition focuses on Islamic inheritance reasoning (ʿIlm al-Mawārīth) through realistic end-to-end problem solving. 

In the previous edition (QIAS 2025), LLMs were evaluated using MCQs on Islamic inheritance reasoning and general Islamic knowledge. Models were required to select one correct answer among six options, without being assessed on the validity of their intermediate reasoning or the correctness of the legal justifications leading to that choice. QIAS 2026 focuses on a single, more challenging task: Islamic inheritance reasoning. Unlike previous MCQ-based evaluations, this task requires models to perform end-to-end reasoning, explicitly generating intermediate steps, applying jurisprudential rules, and computing final inheritance shares. 

End-to-end Islamic inheritance problem solving from natural language scenarios: Given a natural language description of an estate and surviving relatives, systems must identify eligible heirs, apply the correct juristic rules, and compute the exact numerical shares according to Islamic inheritance law.

QIAS 2026 places special emphasis on supporting the development of open-source Arabic models, while still being open to all approaches and model types (commercial and open-source). 

QIAS 2026 is structured into two difficulty levels: 

1. Basic level: where models are required to correctly identify eligible and ineligible heirs and apply the fundamental inheritance rules. 

2. Advanced level: which includes complex jurisprudential cases such as ʿawl, radd 

We provide a complete baseline implementation. This includes code to perform predictions using Fanar API and Gemini, using prompting techniques only (no fine-tuning). 

📂 The baseline code is available here: GitLab

Participants are free to build upon this baseline and explore a variety of approaches, including:

• 🧠 Prompting strategies (zero-shot, few-shot, etc.)

• 🔧 Fine-tuning with the provided training data

For more updates or inquiries, join the QIAS Google group:

Google Group: https://groups.google.com/g/qias2025_shared_task

• December 21, 2025: Release of training, validation, and evaluation scripts

• March 10, 2026: Test Set published

• March 15, 2026: Final results released

• March 25, 2026: System description paper submissions due

• April 25, 2026: Camera-ready versions due

• May 11-16, 2026: LREC 2026 

• Abdessalam BOUCHEKIF (Hamad Bin Khalifa University, Qatar)

• Samer RASHWANI (Hamad Bin Khalifa University, Qatar)

• Mohammed GHALY (Hamad Bin Khalifa University, Qatar)

• Mutaz A. AL-KHATIB (Hamad Bin Khalifa University, Qatar)

• Emad Mohamed  (Nazarbayev University, Kazakhstan )