Saiga Llama 3 8B

Saiga Llama 3 8B is an 8-billion-parameter instruction-tuned model built on Meta's Llama-3-8B-Instruct and optimized for Russian. Developer Ilya Gusev fine-tuned it through several iterations using Russian supervised fine-tuning datasets and preference optimization (KTO), producing a conversational assistant that responds naturally in Russian.

It keeps the native Llama 3 prompt format and ships as open weights on Hugging Face under the Llama 3 Community License, so commercial use is subject to Meta's license terms. The Saiga line has a long history in the Russian NLP community, and this Llama 3 generation is its most capable Russian-first chat model.