Apollo AI Model Targets Ancient Greek Papyrus Restoration
Original: A New Chatbot Wants to Unlock the Secrets in Tattered Ancient Greek Records
Why This Matters
Apollo shows how domain-specific LLMs trained on niche corpora can unlock bottlenecks in specialized scholarship at scale.
The Austrian Academy of Science, partnering with Mistral and Sail Reply, will release Apollo on September 22, 2026 — a large language model trained on 600 million Ancient Greek words from manuscripts, papyri, and inscriptions — to help scholars fill gaps in damaged historical texts.
Academic libraries worldwide hold hundreds of thousands of damaged Ancient Greek papyrus fragments. Restoring them has traditionally required rare specialists who can parse word divisions (Ancient Greek has no spaces), date documents, weigh historical context, and select plausible missing words. Apollo compresses all that expertise into a freely accessible chatbot.
Developed by the Austrian Academy of Science with French AI lab Mistral and technology firm Sail Reply, Apollo is billed as 'the world's first advanced large language model for Ancient Greek.' Its training corpus spans roughly 600 million words from manuscripts, papyri, and inscriptions. The model adapts its register automatically — Homeric Greek for Homer, Doric dialect for Doric inscriptions.
Historian and papyrologist Anna Dolganov describes that adaptability as a key feature. Sail Reply's Dimitris Vlitas told WIRED the capability 'was unthinkable a year ago.' Oxford professor Armand D'Angour, whose university holds the world's largest ancient papyrus collection, called it 'very exciting,' noting that a shortlist of candidate gap-fillers would 'speed up matters considerably.'
Scholars are measured in their expectations. UCL classics professor Stephen Colvin warns against expecting lost Sophocles plays — most unrestored papyri are mundane records like personal letters or civil contracts. The realistic payoff is faster access to granular details about daily life in antiquity and firmer evidence for existing scholarly assumptions.