Ask now has an effort control for supported Claude models. You can choose an effort level for a question and later see that choice in the saved answer’s response details.
The available levels follow the configured model and provider. Medium is the default on supported models; some models also offer Extra high and Max. The control appears on Anthropic and LiteLLM routes with a supported Claude model, with the choices that model accepts.
The chosen effort stays fixed while the answer runs. Each model call in that answer receives the setting, and higher levels allow more room for thinking and answer text within the per-call output ceiling. The setting is not a total spending limit for an investigation that makes several calls.
The browser remembers the preference separately for each signed-in person. If the configured model changes and the selected level is unavailable, Ask refreshes the choices and preserves the unsent question for review. The saved preference remains available when the configured model supports it again.
When an answer needs fresh customer analysis, its continuation carries the original effort choice after that analysis finishes. Customer-analysis stages keep their own settings. Completed, failed and stopped answers retain the requested effort in their saved metadata, so response details can identify the setting used for that attempt.
The control is part of the Ask composer when the configured model supports it.
Development record: November 3, 2026. Written retrospectively.