Austrian Academy, Mistral and Sail Reply Launch Apollo, the First LLM for Ancient Greek

On September 23, 2026, the Austrian Academy of Sciences, French AI lab Mistral, and technology services provider Sail Reply released Apollo, described as the world's first large language model built for ancient Greek. Trained on roughly 600 million ancient Greek words from manuscripts, papyri and stone inscriptions, the model is available free of charge to researchers through a chatbot interface.

Hundreds of thousands of ancient Greek papyrus fragments are held in university libraries worldwide, many too damaged to read. Restoration has traditionally depended on senior scholars manually reconstructing missing text. The Apollo team expects the model to help scholars identify fragments relevant to their research and suggest the most likely words to fill gaps, based on statistical probability.

According to the project team, Apollo adapts its output to the register of the text, using Homeric Greek when it recognizes Homeric works and switching to Doric when handling Doric dialect inscriptions. It offers multiple candidate readings, leaving final decisions to human scholars to reduce the risk of polluting historical records.

Armand D'Angour, professor of classical languages and literature at Oxford, said machine-suggested alternatives could greatly speed up research. The project also suggested the technology could later be adapted to other ancient languages such as Latin and ancient Egyptian, and extended to fields that require indexing large documentary collections.