Apollo AI model accelerates deciphering of ancient Greek papyri

A new AI system called Apollo can fill gaps in millennia-old Greek texts in a matter of hours—a task that previously took specialists years to complete. Developed by the Austrian Archaeological Institute in collaboration with French company Mistral and consultancy Sail Reply, the model was trained on 600 million words extracted from manuscripts, inscriptions, and papyri.

This initiative addresses the challenge posed by approximately one million unread Greek papyri scattered across libraries and museums. Many are torn, erased, or have missing sections, making reconstruction difficult. Apollo suggests the most likely words or phrases to fill in the blanks, enables topic-based searches, and helps interpret handwritten texts. Since Monday, the 22nd, researchers have been able to access the tool for free via a chatbot.

‘There are one million Greek papyri in the world that have never been read, tens of thousands of them in the Austrian National Library’s Papyrus Collection,’ said historian Anna Dolganov, who leads the project. ‘Historical treasures like these are our target.’

The scholar noted that the name Apollo refers to the Greek god of light and patron of the arts and sciences. Heinz Faßmann, President of the Austrian Academy of Sciences, said the tool demonstrates that AI and ancient languages are not mutually exclusive.

From classical theatre to everyday life

Experts caution that most unrestored papyri do not contain lost works by authors such as Sophocles or Homer. Instead, they are everyday documents: letters, marriage contracts, and administrative records. Apollo’s value lies in revealing details of daily life in antiquity and allowing researchers to focus on interpretation rather than manual fragment reconstruction.

Other AI projects in archaeology

Apollo joins a growing list of initiatives using artificial intelligence to recover the past. In 2022, Google DeepMind introduced Ithaca, a system for restoring Greek inscriptions, and in 2025 launched Aeneas, which works with Latin and Greek. In the Vesuvius Challenge, a competition that has already awarded $1.8 million in prizes, scientists used high-resolution tomography and AI to read the complete text of a carbonised Herculaneum scroll for the first time, including 70 columns of a work by philosopher Philodemus on human vices.

A recent study by the University of California, Berkeley, found that the lead in some scroll ink produces a stronger signal in X-ray images, helping identify which documents are most likely to be deciphered.

For the Austrian Academy, Apollo marks only the beginning of a promising journey. ‘This model opens up new perspectives for the study of antiquity,’ says Dolganov.