Module thinking 

Thinking levels and output token budgets.

Constants

MIN_ANSWER_TOKENS

Output tokens kept for the answer when a thinking budget would consume all.

Functions

budget_for_level

The budget for budget-based thinking at level. xhigh and max use high.

budgeted

Budget-based thinking at level on top of the output limit base: the output limit capped to the context, and the budget leaving room to answer under it. Returns (max_tokens, budget).

clamp_level

The nearest supported level: the requested one, else the next higher, else the next lower.

clamp_max_tokens_to_context

Caps max_tokens so the estimated context plus output fits the window.

effort

The requested level clamped to the model, or None when it is off.

requested_max_tokens

pi’s streamSimple output limit: the requested limit, else the model’s, capped to the context window.

supported_levels

The levels a model accepts: only off without reasoning, xhigh and max only when its thinking level map names them, and none the map marks unsupported.