The GigaChat neural network has successfully passed academic certification in the fields of "Electric Power Engineering" and "Thermal Power Engineering" at the Moscow Power Engineering Institute. Sber stated that this is the first Russian language model to pass exams in several engineering specialties at once.
The exam was conducted jointly by Sber specialists, scientists from NIU MEI, and Rosseti experts. To test knowledge, the university prepared educational and scientific materials, as well as a special benchmark for evaluating large language models.
The test included 24 disciplines. The neural network was asked both theoretical and computational questions related to engineering calculations, regulatory framework, and energy economics. Following the certification, GigaChat received a "good" grade.
Sber believes that in the future, the model can be used as an assistant for engineers, designers, teachers, and students of specialized universities, as well as specialists in the digital transformation of energy companies.
This is not the first such exam for GigaChat. Previously, the neural network underwent certifications in more than 20 areas – from medicine and construction to musicology and financial literacy. In addition, the model received international financial analyst certification, exceeding the passing score by more than 10%.
