DeepSeek has released two new artificial intelligence models, V3.2 and V3.2-Speciale, designed to deliver stronger reasoning, more accurate answers, and better support for automated tasks. The update is now available on DeepSeek’s website, app, and API.
The main model, DeepSeek V3.2, focuses on everyday use. It can plan tasks more effectively, make clearer decisions, and use external tools, such as calculators, search functions, or business software, while working through a problem. DeepSeek says it reaches performance similar to the latest premium AI models, but with shorter responses and lower operating costs.
The second model, V3.2-Speciale, is a research version designed for complex problem-solving. It specialises in long, detailed reasoning and advanced mathematics, achieving results at the level of international competitions such as the International Mathematical Olympiad and the International Olympiad in Informatics. However, this model does not support tool use and is available only through a temporary research API.
Both versions build on DeepSeek’s push toward more reliable, agent-style AI—systems that can break down a task into steps, check their work, and perform actions on behalf of the user.
DeepSeek says V3.2 is intended for practical tasks such as writing, analysis, planning, coding, and business automation. Meanwhile, V3.2-Speciale is aimed at universities, developers, and researchers studying advanced reasoning.
Related article: Samsung opens pre-orders for Galaxy Z TriFold in China
