DeepSeek-R1-0528: The new DeepSeek model will bankrupt OpenAI

DeepSeek has released R1-0528, an updated version of their flagship open-source language model that rivals premium closed models like GPT-4, Claude, and Gemini. Trained on 671B tokens, it offers improved reasoning, code generation, and reliability while remaining free and open-source. The model achieves a median benchmark score of 69.45 across MMLU, GSM8K, BBH, and HumanEval tests. Users report strong performance in coding tasks and creative writing, though concerns remain about hallucinations and bias. The model is available through Hugging Face and OpenRouter's free API.