pub const DEFAULT_EFFORT_CAPPED_AT_CEILING: bool = false;Expand description
2026-09-25: Default [behavior].effort_capped_at_ceiling: whether the
qualitative reasoning_effort levels are clamped at E
(max_thinking_budget, or --max-thinking-budget).
false keeps high = 2E and xhigh = 4E, above the ceiling; at the default
E that is 512 and 1024, which
effort_ladder_at_default_ceiling_matches_the_historical_absolutes pins.
true clamps every level at E, for a model whose MODEL.toml records that
a larger budget scores worse (kernels/gb10/qwen3.5-397b-a17b sets it).
The clamp applies only to the effort ladder; an explicit client token
budget is not clamped.