They are all autoregressive. They have just been trained to emit thinking tokens like any other tokens.

reasoning is just more tokens that come out first wrapped in <thinking></thinking>

there are no reasoning LLMs.

This is an interesting denial of reality.

A "reasoning" LLM is just an LLM that's been instructed or trained to start every response with some text wrapped in <BEGIN_REASONING></END_REASONING> or similar. The UI may show or obscure this part. Then when the model decides to give its "real" response, it has all that reasoning text in its context window, helping it generate a better answer.