GigaChat Family: Efficient Russian Language Modeling Through Mixture of Experts Architecture

Generative large language models (LLMs) have become crucial for modern NLP research and applications across various languages. However, the development of foundational models specifically tailored to the Russian language has been limited, primarily due to the significant computational resources required. This paper introduces the GigaChat family of Russian LLMs, available in various sizes, including base models and instruction-tuned versions. We provide a detailed report on the model architecture, pre-training process, and experiments to guide design choices. In addition, we evaluate their performance on Russian and English benchmarks and compare GigaChat with multilingual analogs. The paper presents a system demonstration of the top-performing models accessible via an API, a Telegram bot, and a Web interface. Furthermore, we have released three open GigaChat models in open-source (this https URL), aiming to expand NLP research opportunities and support the development of industrial solutions for the Russian language.

@article{team2025_2506.09440,
  title={ GigaChat Family: Efficient Russian Language Modeling Through Mixture of Experts Architecture },
  author={ GigaChat team and Mamedov Valentin and Evgenii Kosarev and Gregory Leleytner and Ilya Shchuckin and Valeriy Berezovskiy and Daniil Smirnov and Dmitry Kozlov and Sergei Averkiev and Lukyanenko Ivan and Aleksandr Proshunin and Ainur Israfilova and Ivan Baskov and Artem Chervyakov and Emil Shakirov and Mikhail Kolesov and Daria Khomich and Darya Latortseva and Sergei Porkhun and Yury Fedorov and Oleg Kutuzov and Polina Kudriavtseva and Sofiia Soldatova and Kolodin Egor and Stanislav Pyatkin and Dzmitry Menshykh and Grafov Sergei and Eldar Damirov and Karlov Vladimir and Ruslan Gaitukiev and Arkadiy Shatenov and Alena Fenogenova and Nikita Savushkin and Fedor Minkin },
  journal={arXiv preprint arXiv:2506.09440},
  year={ 2025 }
}