GigaChat Family: Efficient Russian Language Modeling Through Mixture of Experts Architecture
by GigaChat team / SaluteDevices
Study a Russian-origin model team’s account of mixture-of-experts architecture, pretraining and evaluation.
The GigaChat team reports its model architecture, pretraining process and experiments used to guide design choices. The paper compares the resulting models on Russian and English benchmarks and describes released model variants.
- Level
- Advanced
- Models
- gigachat
- Category
- Model Training
- Tags
- primary-source, regional-coverage
- Source date
- 2025-06-11
- Added
- 2026-10-18
- Last verified
- 2026-10-18
- Status
- ok
- Prerequisites
- Assumes knowledge of transformer models, mixture-of-experts architectures and language-model training.
- Subcategory
- pretraining
- Region
- RU
- Original language
- en
- Access
- free
- Resource reuse
- open: The inspected paper HTML explicitly states License: CC BY 4.0; this evidence applies to the paper, not automatically to model weights or datasets. Facet definitions
Summary in our own words. Rights remain with the original creator. Credit and correction policy.