DeepSeek launches new AI models with top efficiency and performance

Chinese AI tech company DeepSeek officially launched two new models on Monday evening – DeepSeek-V3.2 and its high-performance variant, DeepSeek-V3.2-Speciale.

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

Competition among leading global tech giants in the AI model space is heating up. In August, OpenAI introduced its flagship model, GPT-5, calling it the smartest and fastest it has ever built. In November, Google unveiled its newest AI system, Gemini-3.0-Pro.

The high-compute DeepSeek-V3.2-Speciale outperforms GPT-5 and shows reasoning skills similar to Gemini-3.0-Pro, according to a report released by the company. The model earned top honors in both the 2025 International Mathematical Olympiad and the International Olympiad in Informatics.

A key technical breakthrough behind this is the DeepSeek Sparse Attention mechanism, which significantly reduces computational complexity while preserving model performance in long-context scenarios.

Founded in July 2023, DeepSeek specializes in researching and developing large language models and multimodal AI technologies.