GigaChat 3.5 Test Drive: Evaluating Model Capabilities for Business Tasks

GigaChat 3.5, released in July and made available with open weights in September, has undergone extensive testing. The authors conducted a comparative analysis of the model's performance in real-world business cases, using a proprietary benchmark to evaluate AI agent operations. The study compared GigaChat 3.5 against current Russian and Chinese alternatives to determine its suitability for enterprise applications. The 'AI agent races' provide insights into the competitiveness of this domestic model against global trends in large language models. The detailed findings regarding performance, accuracy, and efficiency in executing specific tasks are available in the full article.