I've been using Deepseek R1 heavily since it came out. My review is, it is definitely Chinese.

I don't mean any CCCP related stuff. I mean in the way that no matter what the specs on the boxes say, the Chinese thing is just never as nice as the American or European one. You can't point to any specific problem unless you zoom way in, but you just can't miss the quality difference device in hand.

Tests say R1 is better, but my experience is that Claude 3.5 Sonnet gets me a working answer in fewer prompts consistently. I'll be switching back to Claude as my first choice.

I use it mainly for tech and dev related things, so your experience may very.

One neat trick I learned from this experience. I'm using these models through nostr:nprofile1qqsrgyzdaheuckfksqkgxz9r6qys72zh6j46f69hyzkv4a4j4vzfj6gpr9mhxue69uhhyetvv9ujuumwdae8gtnnda3kjctv9ul0q43l so I can hit a drop down and change models mid conversation. Then I prompt "please review the conversation history and tell me what the previous LLM is getting wrong that this isn't working"

HTH HAND.