What changed
Hugging Face announced the release of Ayb 0.5, a family of open-source language models built to advance Armenian-language AI. The release includes three models of different sizes: Ayb-0.8B, Ayb-4B, and Ayb-9B, each tailored for various applications ranging from lightweight text processing to advanced conversations and multimodal understanding.
What this means in practice
The Ayb 0.5 models are based on the Qwen3.5 architecture and were developed through an end-to-end Armenian language adaptation pipeline. Ayb-0.8B is suitable for resource-constrained environments and lightweight Armenian text processing, while Ayb-4B offers a balance between performance and computational efficiency for everyday applications. Ayb-9B, the largest model in the family, achieves the highest ArmBench score and supports multi-turn conversations, complex instructions, and tool calling. The models are available for use by researchers, developers, businesses, and the wider Armenian community.
Limitations and availability
The models were primarily trained on Eastern Armenian, so responses may be less reliable in Western Armenian, regional dialects, and Armenian written in Latin script. They can generate incorrect details and have uneven coverage of recent events. Ayb has limited ability to reason through complex, multi-step problems in thinking mode, and the reported evaluations were conducted with thinking mode disabled. Preference tuning also remains an area for improvement, particularly when multiple responses are technically correct but differ in clarity or usefulness. The models are available on Hugging Face's platform for download and use.
