Orion-14B is a family of open large language models from OrionStarAI, built around a 14-billion-parameter multilingual foundation model. Alongside the base model, the family includes a chat model, a long-context model, a quantized model, a RAG fine-tuned model and an agent fine-tuned model, so developers can pick a variant matched to their use case. The project is hosted on GitHub, where it has earned roughly 810 stars, and the weights are published on Hugging Face and ModelScope.
Released when capable mid-size open models were still scarce, Orion-14B aimed to deliver strong multilingual performance โ with particular depth in Chinese, Japanese and Korean alongside English โ at a size that organizations can actually self-host. The base model was pretrained on a large multilingual corpus, and the derived models extend it in the directions practitioners most often need: dialogue, longer documents, cheaper inference through quantization, retrieval-augmented generation and tool-using agent behavior. Model cards, evaluation results and usage examples are provided in the repository, and the series became a popular base for community fine-tunes in the Chinese open-source ecosystem. The family illustrates a practical release strategy for open models: publish one strong base, then ship the specialized derivatives that most users would otherwise have to fine-tune themselves.
Orion-14B is free and open source under the Apache License 2.0. You download the weights and run them on your own infrastructure at no licensing cost.
Orion-14B is aimed at developers and research teams who need a self-hostable, multilingual LLM โ especially for Chinese-language chat, retrieval or agent applications โ without depending on a commercial API.