DeepSeek launches new AI models with top efficiency and performance

HANGZHOU - Chinese AI tech firm DeepSeek officially released two new models on Monday evening - DeepSeek-V3.2 and its high-compute variant, DeepSeek-V3.2-Speciale.

By implementing a robust reinforcement learning protocol and scaling post-training computation, DeepSeek-V3.2 performs comparably to GPT-5, according to the company. The AI model balances high computational efficiency while maintaining superior reasoning and agent performance.

Competition among leading global tech companies in the AI model arena is intensifying. In August, OpenAI launched its flagship model, GPT-5, describing it as its smartest and fastest model to date. And in November, Google released its latest AI system, Gemini-3.0-Pro.

The high-compute DeepSeek-V3.2-Speciale surpasses GPT-5 and exhibits reasoning proficiency comparable to Gemini-3.0-Pro, according to an essay that was released by the company. The variant achieved gold-medal performance in both the 2025 International Mathematical Olympiad and the International Olympiad in Informatics.

A key technical breakthrough behind is the DeepSeek Sparse Attention mechanism, which substantially reduces computational complexity while preserving 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.