- DeepSeek unveiled two new AI model versions enhancing reasoning and autonomous action capabilities
- DeepSeek-V3.2 matches OpenAI's GPT-5 performance on multiple reasoning benchmarks
- DeepSeek-V3.2 integrates human-like reasoning with tool use such as search and code execution
China's DeepSeek unveiled two new versions of an experimental artificial-intelligence model it released weeks ago, adding fresh capabilities the startup said would help with combining reasoning and executing certain actions autonomously.
While the model the Hangzhou-based company released in September was called DeepSeek-V3.2-Exp in a nod to its experimental status, the latest version is dubbed simply DeepSeek-V3.2. The startup says the new service matches the performance of OpenAI Inc.'s flagship GPT-5 across multiple reasoning benchmarks, suggesting China's open-source systems remain competitive with Silicon Valley's frontier proprietary models on at least certain metrics.
DeepSeek also said the V3.2 version combines that ability to mimic something like human reasoning with the capability to use tools like search engines, calculators and code executors.
“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 post on X.
DeepSeek, which stunned the tech world in January with a breakthrough AI model, had earlier described the experimental version of the model as a step toward next-generation artificial intelligence.
🚀 Launching DeepSeek-V3.2 & DeepSeek-V3.2-Speciale — Reasoning-first models built for agents!
December 1, 2025
🔹 DeepSeek-V3.2: Official successor to V3.2-Exp. Now live on App, Web & API.
🔹 DeepSeek-V3.2-Speciale: Pushing the boundaries of reasoning capabilities. API-only for now.
📄 Tech… pic.twitter.com/SC49UdmCZvThe second version DeepSeek released, V3.2-Speciale, focuses on mathematical computations and other long-thinking capabilities. The startup said the goal is to “push the inference capabilities of open-source models to their limits and explore the boundaries of model capabilities.”
DeepSeek-V3.2-Speciale matches the performance of Google's latest Gemini-3 Pro, and performed at gold medal levels on standardized tests like the International Math Olympiad and International Olympiad on Informatics, the company said.
DeepSeek said it has created a new way to train AI agents, or software programs that are designed to act independently to achieve goals by interacting with their environment, analyzing data, and making decisions without constant human input.
It's a new sign the influential artificial intelligence lab is doubling down on its research momentum with its latest release to make AI processing faster and more efficient. After its breakthrough model upended the AI world in January, the startup is now pushing ahead with fresh innovations to cement its edge in China's AI race. Just last week, the company had released a new open model called DeepSeekMath-V2 which showed strong theorem-proving capabilities in mathematics.
“DeepSeek-V3.2 achieves similar performance with Kimi-k2-thinking and GPT-5 across multiple reasoning benchmarks,” the startup said in an accompanying technical report titled, ‘DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models'.
Watch LIVE TV, Get Stock Market Updates, Top Business, IPO and Latest News on NDTV Profit.
