Not even GPT 5.6 approaches it, why would DeepSeek with less investment?