Translation. Region: Russian Federation –
Source: People's Republic of China in Russian – People's Republic of China in Russian –
An important disclaimer is at the bottom of this article.
Source: People's Republic of China – State Council News
HANGZHOU, Dec. 2 (Xinhua) — Chinese artificial intelligence (AI) technology company DeepSeek officially released two new models on Monday evening, namely DeepSeek-V3.2 and its high-performance version DeepSeek-V3.2-Speciale.
According to the report, the implementation of a robust reinforcement learning protocol combined with scalable post-training computation enabled DeepSeek-V3.2 to demonstrate performance comparable to GPT-5. The AI model strikes a balance between high computational efficiency and the preservation of superior reasoning logic and intelligent agent performance.
Competition among the world's leading tech companies in the field of AI model development is intensifying. In August of this year, OpenAI unveiled its flagship model, GPT-5, which the organization called the most advanced and responsive model to date. In November of this year, Google unveiled its latest AI system, dubbed Gemini-3.0-Pro.
Remarkably, the high-performance DeepSeek-V3.2-Speciale variant not only outperforms GPT-5 but also matches Gemini-3.0-Pro in reasoning and logical inference, according to a recent company publication. Furthermore, it won gold medals at both the 2025 International Mathematical Olympiad (IMO) and the International Olympiad in Informatics (IOI).
A significant technical advancement is the DeepSeek Sparse Attention mechanism, which results in a significant reduction in computational complexity in long-term context scenarios while maintaining the performance of the AI model.
DeepSeek was founded in July 2023 with a primary focus on research and development of large-scale language models and multimodal AI technologies. -0-
Please note: This information is raw content obtained directly from the source. It represents an accurate account of the source's assertions and does not necessarily reflect the position of MIL-OSI or its clients.