I was praising Microsoft's Phi4 the other day and you know what I still praise it. It is a very good Chat LLM. Very much.
DeepSeek R1 is not a Chat LLM, it is more of an agent or worker. You can still discuss with it, but it will print out a lot of detail even what it is thinking. They categorize this model, like GPT o1 and o3, as Reasoning LLMs.
Anything you type into Reasoning LLMs will cause these LLMs to output a lot of details because it is trying to figure out how to answer your question with as much detail as possible.
The way to approach Reasoning LLM is to give it a prescribed goal, constraints or context, negative considerations (technically part of context) and any more detail such as examples and output examples to help the LLM with getting the best result as possible.
It's crazy how the Chinese created a Reasoning LLM that fits on your computer and performs more complex tasks than previous open LLMs. USA has it's work cut out for it. There's no way ChatGPT's model would fit on non server computers, but god damn. DeepSeek is the new Open AI, they actually made this open source for everyone to use.
DeepSeek R1 is available on Ollama (https://ollama.com/library/deepseek-r1) and of course available on HuggingFace.