DeepSeek, dubbed the “God of Open Source,” has unveiled its latest model, DeepSeek V3.2, which dramatically outperforms GPT-5 High and rivals Google’s Gemini 3.0 Pro. With achievements including four gold medals in prestigious international Olympiads, V3.2 utilizes innovative DSA sparse attention architecture, successfully breaking the “impossible triangle” of speed, cost, and intelligence.
This model excels in various reasoning benchmarks, demonstrating superior capabilities in mathematics and programming. Two versions are available: DeepSeek V3.2, designed for daily use, and DeepSeek V3.2-Speciale, which focuses on advanced reasoning but is accessible via API.
Key innovations include the seamless integration of thinking and tool use, enhanced training environments, and reduced computational costs—allowing for both rapid and intelligent responses. DeepSeek V3.2 is a significant leap in open-source AI, showcasing smarter architecture and refined training methodologies to push the boundaries of reasoning. Explore the models on HuggingFace and ModelScope.
Source link
