Outlook Business Desk
DeepSeek has announced its new V3.2 series as it attempts to regain momentum in the global AI race, where Google’s Gemini and OpenAI’s GPT models currently dominate performance.
DeepSeek rose to prominence early this year when its R1 and V3 models delivered ChatGPT-level performance at far lower development costs, positioning the Chinese startup as a serious challenger to Western AI dominance.
However, DeepSeek struggled to keep pace in recent months while Google expanded its Gemini and Nano portfolio and OpenAI advanced with agentic tools and major model upgrades like GPT 5.1.
The new V3.2 and V3.2 Speciale models mark DeepSeek’s return, with the company positioning them as reasoning-first systems specifically designed for agent-driven tasks and complex problem-solving.
DeepSeek says the V3.2 model delivers general performance comparable to GPT 5, while the V3.2 Speciale reportedly matches the reasoning abilities of Google’s latest Gemini 3 Pro model.
DeepSeek’s V3.2 Speciale secured gold-level scores in the 2025 International Mathematical Olympiad and International Olympiad in Informatics, areas previously dominated by leading Google and OpenAI systems.
DeepSeek says V3.2 Speciale surpasses GPT 5 High and Gemini 3 Pro on strict mathematics tests like AIME and HMMT and leads coding logic evaluations, including top rankings on CodeForces.
DeepSeek V3.2 replaces the earlier experimental version as the default model on its website and app, while the more advanced V3.2 Speciale is currently available only through API access.