DeepSeek V4 Shocks AI World: Open-Source Model Outperforms GPT-5 on Key Benchmark

A new open-source AI model, DeepSeek V4, has taken the tech world by storm, achieving an 88.1% score on the MMLU benchmark and surpassing GPT-5's reported performance in general knowledge. Developed for under $6 million, DeepSeek V4 is fully open-weight, enabling local deployment, and boasts a 256K token context window. It quickly became the most-downloaded model on HuggingFace, signaling a major shift in the competitive landscape. This release marks a significant milestone for open-source AI, posing a serious challenge to proprietary models like those from OpenAI.