> to conclude if it is thinking or not I would want to look inside its head

https://arxiv.org/abs/2607.03502

a non-thinking token model (just "completion") can answer one-step questions but generally not multistep questions. however, if you append [n] of a single token (e.g. period, space), it is able to use the activations in the higher layers of the blank tokens as a "scratchpad" to seemingly work through the complex question through "causal token time" and deliver a correct answer

if you wanted to further study the phenomenon you could probably run the experiment again, and the ablate or corrupt those intermediate activations to get a feel for what it was thinking at the "time".