Alexandria-based artificial intelligence startup TokenAI has released two new multimodal models, Horus Hiero and Horus Hiero Mini, built to read, translate and reason across Ancient Egyptian hieroglyphics, modern Arabic dialects and more than 100 other languages.
The company says the models are the first in the Arab world to include a native hieroglyphic processing engine trained on epigraphic drawings, stone reliefs and papyrus documents. Both were developed without external funding and will be released as open-weight under a custom developer license, with a Horus Chat web and mobile interface to follow.
TokenAI, founded by Egyptian entrepreneur Assem Sabry, positions the release as its most significant so far this year. It follows Horus 1.0-4B, a four billion parameter open-source model released in April under an MIT license that the developer said outperformed Meta’s Llama 3.1-8B and Google’s Gemma-2-9B on the MMLU benchmark despite having fewer parameters.
The flagship Horus Hiero 9B and compact Horus Hiero Mini 4B are optimized for the cultural heritage, languages and dialects of Egypt and the broader Middle East and North Africa region. Both feature a 128,000-token context window, long enough to process entire books or multiple papyrus documents in a single inference pass, and support Arabic dialect mapping across colloquial Egyptian, Levantine, Gulf and North African variants.
The core capability is native hieroglyphic processing. Where existing vision models can identify hieroglyphic symbols but struggle to parse ancient grammar or produce structured translations from visual inputs, TokenAI says its engine enables direct visual translation of ancient inscriptions into structured English and Arabic without relying on external vision models.
To measure that performance, TokenAI has also released its own Hieroglyphic AI Benchmark, covering four dimensions of ancient text processing: recognition, or identifying and segmenting individual symbols per Gardiner’s Sign List; transliteration, converting symbols into standardized phonetic representations; translation, producing linguistically and contextually accurate renderings in modern languages; and contextual reasoning, understanding religious, historical and cultural connections between symbols. The company gives the example of a combination of the Ankh, Eye of Horus, Was-sceptre and Ma’at symbols expressing a sacred meaning about the safety of the individual and the pharaohs’ relationship with the gods, rather than a sequence of separate words.
According to TokenAI’s internal evaluation, Horus Hiero scores 90 percent overall on the benchmark, with 92.4 percent on visual symbol recognition and 89.3 percent on grammatical translation. Horus Hiero Mini scores 84.2 percent on the same benchmark and is designed to run on standard CPUs and mobile devices, making offline real-time hieroglyphic translation feasible on budget hardware.
On broader benchmarks, the company says the flagship 9B model scores 79.3 percent on MMLU-Pro, 78.1 percent on GPQA Diamond and 83.7 percent on HumanEval for Python coding. The Mini scores 74.2 percent on MMLU-Pro and 72.3 percent on GPQA Diamond.
Practical use cases outlined by the developer include real-time camera-based inscription translation for tourists and museum visitors, multi-dialect enterprise document search, academic Egyptology tools and high-fidelity OCR of historical archives.
TokenAI has also built Horus Lens 1.0, a text-to-image and image-to-image generation model, and is developing a more powerful successor to Horus Hiero, alongside the forthcoming Horus Chat interface.
