TokenAI's Horus Hiero decodes ancient Egyptian hieroglyphs with open-source AI
First multimodal model for hieroglyph translation runs on CPUs and mobile devices
TokenAI has released Horus Hiero, the first open-source AI model family dedicated to translating Ancient Egyptian hieroglyphs. Built on Qwen 3.5, the models are available in two sizes: Horus Hiero 9B and the lighter Horus Hiero Mini 4B, which is optimized for CPUs and mobile devices. The models are truly multimodal, processing text, images, and video input, and support approximately 150 languages for translation. This combination makes Horus Hiero uniquely positioned for archaeology, tourism, and education sectors seeking to make hieroglyphic texts accessible without specialized expertise.
The models also demonstrate strong general reasoning and coding performance: 79% on MMLU-Pro, 63% on LiveCodeBench, and 84% on HumanEval. Horus Hiero offers a 512K token context window, expandable up to 1M tokens, one of the largest in the Arab AI ecosystem. TokenAI has released the models fully open source on Hugging Face, with integration support via the NeuralNode framework. This release aims to democratize access to Ancient Egyptian heritage, boost cultural tourism, and encourage broader study of hieroglyphic writing systems.
- Two variants: 9B flagship and 4B Mini optimized for CPU/mobile inference
- Multimodal support for text, images, and video across ~150 languages
- 512K context window (expandable to 1M tokens); benchmarks include 84% HumanEval
- Fully open source on Hugging Face with NeuralNode framework support
Why It Matters
Breaks down language barriers for archaeology and tourism while advancing multimodal translation capabilities.