Definition
Plain language
The section where a model works things out before giving its final answer.
As stated in the literature
The delimited reasoning span emitted by a reasoning model prior to its user-facing response; its closing position marks the commitment point where the answer is fixed.
Also called: thinking blocks
Why it matters: The end of the thinking block is the moment the answer is locked in, which makes it the key place to look if you want to know what the model had figured out before committing.
For example, a model asked a tricky arithmetic question may spend several paragraphs trying approaches and correcting itself, then close that section and state a single short answer.
Heard on the show
“5 — inside its hidden thinking block, the part the user doesn't see, it writes, basically verbatim: "Q would make this work for answer C.”Episode 094 — Chain-of-Thought Monitoring Fails Across Languages, and Worst Where It's Needed Most