I just tested Deepseek R1 a little, and it spend half its "thinking time" on the formatting of its citation, how to cite if it did use websearch and thus has no links, etc. Maybe this just occasionally happens, but it may also be possible to fix the bug with small tweaks to the prompts setups.
Disregarding the fact that my prompt would not actually require a reasoning model in the first place, this was the thread: https://kagi.com/assistant/aa115980-f3da-49bb-ad38-248fa13a0dc0