DeepSeek just dropped two insanely powerful AI models that rival GPT-5 and they're totally free

DeepSeek, a Chinese AI startup, released DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, AI models claiming parity with OpenAI's GPT-5 and Google's Gemini-3.0-Pro. DeepSeek-V3.2-Speciale achieved gold medal performances in international academic competitions, showcasing its elite reasoning abilities. The models utilize "DeepSeek Sparse Attention," reducing computation costs significantly, especially for long inputs. The standard V3.2 model performs well on reasoning and coding tasks, exceeding GPT-5 on some benchmarks. DeepSeek's models are open-source under an MIT license, differing from the proprietary approach of US competitors. The models are trained to think while using tools, enhancing multi-step problem solving capabilities. DeepSeek faces regulatory challenges, with some European and US authorities raising data security concerns. The company indicates it can utilize Chinese-made chips, potentially circumventing US export controls. DeepSeek’s release challenges the notion that AI leadership requires massive expenditures. The company acknowledges limitations in world knowledge but plans to address this. DeepSeek's advancements signal a new phase in the AI race, with open-source models challenging American dominance.