On December 1, 2025, DeepSeek proudly announced the launch of two official model editions: DeepSeek-V3.2 and DeepSeek-V3.2-Speciale. The official website, mobile application, and API have all been seamlessly upgraded to incorporate the official iteration of DeepSeek-V3.2. Presently, the Speciale edition is exclusively accessible as a provisional API service, intended for community evaluation and scholarly research purposes. DeepSeek-V3.2 is meticulously engineered to strike an optimal balance between reasoning prowess and output length, rendering it an ideal choice for Q&A sessions and a wide array of Agent task scenarios. In contrast, DeepSeek-V3.2-Speciale stands as its augmented long-form thinking variant, integrating the theorem-proving capabilities of DeepSeek-Math-V2, with a visionary goal of pushing the boundaries of model capabilities to new frontiers.
