🇪🇬 The First Open-Source AI Model in Egypt!
The 4B-parameter model beats larger rivals like Llama 3.1 8B on key reasoning benchmarks.
TokenAI has officially launched the Horus-1.0 series, marking Egypt's first fully open-source AI model trained from scratch. The initial release, Horus-1.0-4B, is a 4-billion parameter model with an 8K context window, available in seven variants to suit different hardware constraints. It is distributed under an open-source license and integrated into the neuralnode Python framework, which also provides access to Replica's text-to-speech with 20 voices across 10 languages, including Arabic.
The model's significance is underscored by its performance, where it reportedly outperforms established models like Qwen 3.5-4B and, notably, the larger Llama 3.1 8B on challenging reasoning benchmarks such as MMLU Pro. The project aims to build a foundational Egyptian AI infrastructure, positioning Horus as a contender for the strongest language model in the Arab world and a milestone for the region's open-source AI development.
- First fully open-source, from-scratch trained LLM from Egypt, with 4B parameters and 8K context.
- Outperforms larger models like Llama 3.1 8B on MMLU Pro reasoning benchmarks.
- Available in 7 compressed versions via neuralnode framework, with integrated multilingual TTS.
Why It Matters
Establishes a competitive, open-source AI foothold for the Arab world, enabling local development and reducing dependency on foreign models.