DeepSeek says it has built two new AI models that can match or beat the most advanced systems from OpenAI and Google, marking another escalation in the global race for high-level machine intelligence.
In a fresh update on X, the company introduces DeepSeek V3.2 and a more powerful version called V3.2 Speciale, positioning them as direct challengers to Chat GPT-5 and Google’s Gemini 3 family.
DeepSeek shares data showing that their models are on par with GPT-5 and Gemini 3 in terms of reasoning.
Reasoning is what separates a chatbot that answers questions from an AI that can think through multi-step tasks. It is the difference between an assistant that gives suggestions and an agent that can make plans, troubleshoot, analyze documents, run tools and finish work the way a junior employee would.
DeepSeek says the regular V3.2 model is meant for daily use, while the stronger version, V3.2 Speciale, is designed for complex reasoning tasks and is currently available only through its API.
“World-Leading Reasoning:
- V3.2: Balanced inference vs. length. Your daily driver at GPT-5 level performance.
- V3.2-Speciale: Maxed-out reasoning capabilities. Rivals Gemini-3.0-Pro.
- Gold-Medal Performance: V3.2-Speciale attains gold-level results in IMO, CMO, ICPC World Finals & IOI 2025.”
DeepSeek’s data suggests that the V3.2-Speciale model can break apart hard problems, track long chains of logic and generate correct solutions with fewer errors. This is the type of capability needed for agents, the next stage of AI where systems handle real work rather than chat.
DeepSeek says it trained these models using a massive synthetic dataset built to teach agents how to think and act in realistic environments.
The company says the training process covered more than 1,800 simulated environments and 85,000 complex instructions. The focus was not just on conversation, but helping the model understand how to use tools, plan actions and operate step by step like a digital worker.
“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.”
DeepSeek’s model advances come as China has overtaken the US in the number of AI model downloads, suggesting AI leadership is moving away from Silicon Valley. In October, Nvidia CEO Jensen Huang warned that China is winning the race in AI adoption and that US firms could fall behind if they focus too heavily on model development instead of deployment.
Disclaimer: Opinions expressed at CapitalAI Daily are not investment advice. Investors should do their own due diligence before making any decisions involving securities, cryptocurrencies, or digital assets. Your transfers and trades are at your own risk, and any losses you may incur are your responsibility. CapitalAI Daily does not recommend the buying or selling of any assets, nor is CapitalAI Daily an investment advisor. See our Editorial Standards and Terms of Use.