Deepseek R1 is genuinely terrible. If you think it isn’t you should try it out before ceding to the hype. Llama is better. It could be that the way I use those models biases me, as I primarily want a model that gets me what I want fast, rather than accurately.