The official version of DeepSeek V3.2 is released: strengthening agent capabilities and integrating thinking and reasoning

Today, we are releasing two official models simultaneously: DeepSeek-V3.2 and DeepSeek-V3.2-Speciale. DeepSeek-V3.2 is our first model that integrates thinking into tool use, and supports tool calls in both thinking and non-thinking modes. We propose a large-scale agent training data synthesis method, which constructs a large number of "difficult-to-answer, easy-to-verify" reinforcement learning tasks (1800 + environments, 85,000 + complex instructions), which greatly improves the generalization ability of the model. (DeepSeek)