China’s rising AI startup DeepSeek has announced two upgraded versions of its experimental artificial intelligence model, adding powerful new abilities aimed at strengthening the system’s reasoning performance and its capacity to autonomously execute tasks.
Just weeks after releasing its experimental model DeepSeek-V3.2-Exp, the Hangzhou-based company has now introduced a more advanced model simply named DeepSeek-V3.2. According to the startup, this updated version delivers performance that matches OpenAI’s GPT-5 on several major reasoning benchmarks, indicating that China’s open-source AI ecosystem continues to remain highly competitive with Silicon Valley’s most cutting-edge proprietary models.
One of the most notable upgrades is the model’s ability to seamlessly blend human-like reasoning with advanced tool-use capabilities. DeepSeek said the new system can directly integrate its “thinking” processes with tools such as search engines, calculators, and code execution frameworks — an important feature that brings AI closer to autonomous multi-step task completion.
“DeepSeek-V3.2 is our first model to integrate thinking directly into tool-use, and also supports tool-use in both thinking and non-thinking modes,” the company said in a statement posted on X.
DeepSeek captured global attention earlier this year after unveiling a groundbreaking AI model in January, widely regarded as one of the strongest open-source releases to emerge from China. The company described its earlier experimental version as an essential step toward developing more sophisticated, next-generation artificial intelligence systems.
With these latest releases, DeepSeek continues to position itself as a formidable challenger in the global AI race, bridging the gap between open-source innovation and cutting-edge reasoning performance.