DeepSeek Launches V3.2 Models with Integrated Reasoning and Tool Use
Chinese AI company DeepSeek has officially released its new V3.2 and V3.2-Speciale large language models, available across web platforms, mobile applications, and APIs. The standard V3.2 model introduces built-in reasoning capabilities specifically designed for agent tasks and marks a significant milestone as the company's first model to support tool calls in both reasoning and non-reasoning modes. DeepSeek claims that V3.2 achieves performance comparable to GPT-5 on public reasoning benchmarks while simultaneously reducing output length and computational costs, making it highly optimized for daily operational tasks. Additionally, the specialized V3.2-Speciale variant is currently accessible via a temporary API for research purposes. This version focuses on extending long-form reasoning and mathematical proof capabilities, having reportedly achieved gold-level results in prestigious competitions such as the IMO, CMO, ICPC World Finals, and IOI 2025. However, unlike the standard version, the Speciale model consumes more tokens and does not support tool use. This launch highlights DeepSeek's continued advancement in efficient, high-performance AI reasoning technologies.
Wire timeline
DeepSeek Launches V3.2 Models with Integrated Reasoning and Tool Use
Chinese AI company DeepSeek has officially released its new V3.2 and V3.2-Speciale large language models, available across web platforms, mobile applications, and APIs. The standard V3.2 model introduces built-in reasoning capabilities specifically designed for agent tasks and marks a significant milestone as the company's first model to support tool calls in both reasoning and non-reasoning modes. DeepSeek claims that V3.2 achieves performance comparable to GPT-5 on public reasoning benchmarks while simultaneously reducing output length and computational costs, making it highly optimized for daily operational tasks. Additionally, the specialized V3.2-Speciale variant is currently accessible via a temporary API for research purposes. This version focuses on extending long-form reasoning and mathematical proof capabilities, having reportedly achieved gold-level results in prestigious competitions such as the IMO, CMO, ICPC World Finals, and IOI 2025. However, unlike the standard version, the Speciale model consumes more tokens and does not support tool use. This launch highlights DeepSeek's continued advancement in efficient, high-performance AI reasoning technologies.
TechNode