Chinese AI start-up DeepSeek has introduced its most advanced model, V3.2-Speciale, claiming it matches Google DeepMind's Gemini 3 Pro in reasoning abilities. Despite limitations in semiconductor access, DeepSeek's model achieved unmatched success by securing gold-medal performance on the International Mathematical Olympiad test, a feat only previously accomplished by undisclosed OpenAI and Google models. While the V3.2 variant is open-source on Hugging Face, V3.2-Speciale is accessible via a restricted API, reflecting its complex performance capabilities.
Chinese AI start-up DeepSeek has introduced its most advanced model, V3.2-Speciale, claiming it matches Google DeepMind's Gemini 3 Pro in reasoning abilities. Despite limitations in semiconductor access, DeepSeek's model achieved unmatched success by securing gold-medal performance on the International Mathematical Olympiad test, a feat only previously accomplished by undisclosed OpenAI and Google models. While the V3.2 variant is open-source on Hugging Face, V3.2-Speciale is accessible via a restricted API, reflecting its complex performance capabilities.