The new models are based on three major breakthroughs such as DeepSeek Sparse Attention, the Scalable Reinforcement Learning Framework, and the Large-Scale Agentic Task Synthesis Pipeline. (Express: Image) Chinese AI startup DeepSeek has released two new AI models, named V3. 2 and VSpeciale.

According to the AI ​​upstart, the new models performed on par with state-of-the-art models like the GPT-5 and Gemini 3 Pro. The models were able to achieve this performance while reducing costs and keeping them accessible under open-source licenses. DeepSeek-v3.

2 reportedly matches or is close to the performance of Cloud Sonnet 4. 5, GPT-5, and Gemini 3 Pro in use cases such as tool usage, coding testing, etc.

Meanwhile, the special model achieved gold medal scores in the 2025 International Mathematics Olympiad and Informatics Olympiad.