Comment by faitswulff

Anthropic's position is that thinking tokens aren't actually faithful to the internal logic that the LLM is using, which may be one reason why they started to exclude them:

https://www.anthropic.com/research/reasoning-models-dont-say...