In a striking development in the global AI race, Chinese company DeepSeek has unveiled the DeepSeek-V3.2 models, claiming they match GPT-5’s logical reasoning performance and Gemini-3 Pro in mathematics. This step confirms that Chinese open-source models remain a formidable competitor to proprietary Silicon Valley models.
Article contents:
- A new challenger from the East
- DeepSeek-V3.2: Reasoning and tool use
- Clash of the giants: Matching GPT-5
- The specialized model: V3.2-Speciale excels in mathematics
- Towards the next generation: AI agents
- Frequently asked questions
A new challenger from the East
The fierce competition in artificial intelligence continues between tech giants in the United States and China. In the latest developments, Chinese startup DeepSeek has revealed two new versions of an experimental AI model, adding new capabilities that the company says will help combine logical reasoning with executing specific actions independently.
The Hangzhou-based company, which made waves in the tech world last January with an advanced model, continues to push the boundaries of innovation to cement its position in this heated global race.
DeepSeek-V3.2: Reasoning and tool use
The company has named its latest release DeepSeek-V3.2. The standout feature in this version is its ability to combine simulated human reasoning with the capability to use external tools such as search engines, calculators, and code interpreters.
The company said in a post on X: “DeepSeek-V3.2 is our first model to directly integrate reasoning into tool use.” This ability to use tools independently is known as “AI Agents,” which are programs designed to work autonomously to achieve specific goals without continuous human intervention.
Clash of the giants: Matching GPT-5
Most impressively, DeepSeek claims that the new V3.2 model matches the performance of OpenAI’s flagship GPT-5 model across multiple logical reasoning benchmarks. This claim, if proven true, indicates that China’s open-source systems remain capable of competing with sophisticated proprietary models in Silicon Valley, at least in certain specific metrics.
In the accompanying technical report titled “DeepSeek-V3.2: Pushing the Boundaries of Open Large Language Models,” the company emphasized achieving performance similar to GPT-5 in logical reasoning areas, placing it among the world’s leading companies.
The specialized model: V3.2-Speciale excels in mathematics
In addition to the main model, DeepSeek released a second version called V3.2-Speciale, which focuses specifically on mathematical calculations and other long-form reasoning capabilities. The startup stated that the goal is “to push the reasoning capabilities of open-source models to their absolute limits.”
The company claims that DeepSeek-V3.2-Speciale matches the performance of Google’s latest Gemini-3 Pro models, achieving gold medal-level results in standardized tests such as the International Mathematical Olympiad and the International Olympiad in Informatics. This focus on mathematical capabilities is particularly important, as it serves as a strong indicator of the model’s advanced logical reasoning abilities.
Towards the next generation: AI agents
DeepSeek also announced that it has devised a new method for training AI agents. These agents are programs designed to operate independently to achieve goals by interacting with their environment, analyzing data, and making decisions without continuous human intervention.
The ability to combine advanced logical reasoning, tool use, and independent action execution are the core components of effective AI agents. These developments indicate that DeepSeek is not only focused on improving language models, but is actively striving to build the next generation of artificial intelligence capable of handling complex real-world tasks. As the race continues, companies like DeepSeek prove that AI innovation is not confined to Silicon Valley, and that China is a key player in shaping the future of this technology.
Frequently asked questions
Q: What is DeepSeek?
An: It is a Chinese startup specializing in the development of advanced, open-source AI models, and is considered a prominent player in the global AI competition.
Q: What is new in the DeepSeek-V3.2 model?
An: The new addition is its claimed ability to match GPT-5’s performance in logical reasoning, alongside its advanced capability to integrate reasoning with the use of external tools like search engines and calculators to execute tasks independently.
Q: Why is it important for these models to be open-source?
An: Being open-source means the code and data are publicly available, which accelerates innovation and allows developers around the world to use and build upon them, posing a challenge to closed proprietary models developed by companies like OpenAI and Google.
Q: What is the specialization of the V3.2-Speciale model?
An: This model specializes in mathematical calculations and long-form reasoning capabilities, achieving performance that rivals Gemini-3 Pro and earning gold medal tiers in the International Mathematical Olympiad.
