Sber’s GigaChat neural network has reportedly outperformed OpenAI’s ChatGPT-3.5 in both English and Russian responses, according to a report from Prime, the economic information agency.
Andrey Belevtsev, the senior vice president of Sberbank and head of the Technology block, explained that the bank’s team completed the Massive Multitask Language Understanding exam. The test covered 57 knowledge areas, spanning mathematics, history, medicine, physics, world knowledge, and problem solving. Belevtsev noted that GigaChat delivered higher quality Russian responses than GPT, underscoring the model’s capabilities in handling nuanced language tasks.
GigaChat entered test mode in April of the previous year and opened to the public in early September. Today, more than 2.5 million users rely on it. The system is capable of handling a wide range of tasks, from carrying on conversations and drafting text or code to answering questions. The ensemble also includes a Kandinsky model, enabling the creation of images, which broadens its practical applications beyond pure text processing.
Belevtsev emphasized that Sber continues to advance GigaChat to meet business needs. He outlined how organizations can engage with the platform through the GigaChat API, a software interface that provides access to Sber’s neural network. The API enables companies to build custom solutions and optimize internal processes, work effectively with large bodies of text, tailor articles to a desired style and format, and assist in research and analysis based on available data.