Artificial intelligence start-up DeepSeek has launched its most advanced model, DeepSeek-V3.2-Speciale, which rivals Google DeepMind’s Gemini 3 Pro in specific tasks, despite limitations in access to advanced semiconductor chips. This announcement, made during the Conference on Neural Information Processing Systems (NeurIPS), has ignited discussions within the AI research community. DeepSeek claims that V3.2-Speciale matches Gemini 3 Pro in reasoning and its base model, V3.2, parallels OpenAI’s GPT-5, released in August. Notably, V3.2-Speciale excelled by achieving gold-medal performance on the International Mathematical Olympiad test, a feat previously accomplished only by internal models from OpenAI and Google DeepMind. While the V3.2 model is open-sourced on Hugging Face, the V3.2-Speciale is only available via an API due to its higher token usage. This development underscores DeepSeek’s commitment to advancing AI technology and contributing to the global AI ecosystem.
Source link
Share
Read more