Chinese artificial intelligence (AI) technology company DeepSeek officially launched two new evening local time, namely DeepSeek-V3.2 and its high-computing variant, DeepSeek-V3.2-Speciale.
By implementing a robust reinforcement learning protocol and leveraging post-training computation, DeepSeek-V3.2 has performance comparable to GPT-5, the company said.
The AI model balances high computational efficiency while maintaining superior reasoning capabilities and agent performance.
Competition among leading global technology companies in the field of AI models is heating up.
In August, OpenAI launched its flagship model, GPT-5, which it described as the smartest and fastest model they had ever developed.
Then in November, Google released its latest AI system, Gemini-3.0-Pro.
The computationally powerful DeepSeek-V3.2-Speciale outperformed GPT-5 and demonstrated reasoning capabilities comparable to Gemini-3.0-Pro, according to a paper released by DeepSeek.
DeepSeek-V3.2-Speciale achieved gold medals in the 2025 International Mathematical Olympiad and the 2025 International Olympiad in Informatics.
The key technical breakthrough behind it is the DeepSeek Sparse Attention mechanism, which significantly reduces computational complexity while maintaining model performance in long-context scenarios.
Founded in July 2023, DeepSeek focuses on the research and development of large language models and multimodal AI technologies.
