ReasoningEffortLevel
Enums · AiDotNet.Enums
How much internal deliberation a reasoning model is asked to spend before it answers.
For Beginners
Some newer AI models “think” privately before replying, and you can tell them how
hard to think. Low is quick and cheap, High is slow and thorough, and
Unspecified simply does not mention the setting so the provider picks for you. If you are unsure,
leave it unspecified and only raise it when the model’s answers are not good enough.
How It Works
Reasoning models (OpenAI’s o-series and GPT-5 families, and the GPT-OSS releases) generate hidden reasoning
tokens before the visible answer. The provider exposes one knob for how many of those it is willing to spend,
sent on the wire as reasoning_effort. More effort usually means a better answer on hard problems and a
slower, more expensive call; less effort is right for mechanical edits where the model already knows what to do.
Unspecified is not a level: it means “send no reasoning_effort field at all” and let the
provider apply its own default. That is the reference OpenEvolve behaviour, which only adds the field when the
configuration supplies one, so leaving this value alone reproduces upstream request bodies exactly.
Fields
| Field | Summary |
|---|---|
ReasoningEffortLevel High |
The most deliberation the model offers, sent as high. |
ReasoningEffortLevel Low |
Fast and cheap deliberation, sent as low. |
ReasoningEffortLevel Medium |
Balanced deliberation, sent as medium. |
ReasoningEffortLevel Minimal |
The smallest amount of deliberation the model offers, sent as minimal. |
ReasoningEffortLevel Unspecified |
No effort field is sent; the provider’s own default applies. |
int value__ |