Chinese Startup DeepSeek Releases Free AI Models That Outperform GPT-5 and Gemini-3.0-Pro
Summary
Chinese AI startup DeepSeek releases free, open-source models DeepSeek-V3.2 and DeepSeek-V3.2-Speciale that reportedly outperform OpenAI's GPT-5 and Google's Gemini-3.0-Pro, achieving gold medals in international competitions while cutting computing costs by 70% through innovative Sparse Attention technology.
Key Points
- Chinese AI startup DeepSeek releases two new models, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, claiming performance that matches or exceeds OpenAI's GPT-5 and Google's Gemini-3.0-Pro while being completely free and open-source
- The models achieve breakthrough performance on elite competitions including gold medals at the 2025 International Mathematical Olympiad and International Olympiad in Informatics, with the Speciale variant scoring 96.0% on AIME 2025 compared to GPT-5's 94.6%
- DeepSeek introduces innovative Sparse Attention technology that reduces computing costs by 70% and enables 'thinking in tool-use' capabilities, allowing continuous reasoning across multiple tool calls while processing documents up to 128,000 tokens