Chinese company DeepSeek has unveiled two new versions of its experimental artificial intelligence model. The startup said the new capabilities will help combine logical thinking and autonomous execution of certain actions.
Chinese company DeepSeek has unveiled two new versions of its experimental artificial intelligence model. The startup said the new capabilities will help combine logical thinking and autonomous execution of certain actions.
The model the company released in September was called DeepSeek-V3.1-Exp, indicating its experimental status. Instead, the latest version has been simply called DeepSeek-V3.2, Bloomberg reports .
The startup claims that the new service is not inferior to OpenAI's flagship GPT-5 in a number of logical thinking tests.
DeepSeek also noted that version V3.2 combines the ability to simulate something like human reasoning with the ability to use tools such as search engines, calculators, and code execution tools.
“DeepSeek-V3.2 is our first model that integrates thinking directly into tool use, and also supports tool use both in and without thinking mode,” the company said.
The second version released by DeepSeek, V3.2-Speciale, focuses on mathematical computation and other long-term thinking capabilities. The startup said the goal is to “push the inference capabilities of open-source models to their limits and explore the limits of the model’s capabilities.”
DeepSeek-V3.2-Speciale matches the performance of Google's latest Gemini-3 Pro model and has achieved gold medal-level results in standardized tests such as the International Mathematical Olympiad and the International Olympiad in Informatics, the company said.
DeepSeek said it has created a new way to train artificial intelligence agents, or software designed to act independently to achieve goals by interacting with the environment, analyzing data and making decisions without constant human intervention.
