Ollama has been deliberately misrepresenting R1 distill models as "R1" for marketing purposes. A lot of "AI" influencers on social media are unabashedly doing the same. Ollama's default "R1" model is a 4-bit RTN quantized 7B model, which is nowhere close to the real R1 (a 671B parameter fp8 MoE).
https://www.reddit.com/r/LocalLLaMA/comments/1i8ifxd/ollama_...