Table of Contents

Enum ReasoningLevel

Namespace
LMKit.TextGeneration.Chat
Assembly
LM-Kit.NET.dll

Controls how much deliberate reasoning a model is encouraged to use. This value is a hint to the model/runtime (not a guarantee) and can be surfaced in prompts or runtime settings to bias response style/latency.

public enum ReasoningLevel

Fields

None = 0

Do not include any explicit “reasoning effort” hint. Lets the model/runtime choose its own level, or use product defaults.

Low = 1

Minimal internal deliberation. Best for simple lookups, short answers, or latency-sensitive paths.

Medium = 2

Balanced speed/quality suitable for most requests that need some multi-step thinking without heavy planning.

High = 3

Increased internal deliberation for complex tasks (planning, tricky logic/math, multi-constraint synthesis). May increase latency and token usage.

Examples

Enabling chain-of-thought reasoning in a conversation:

using LMKit.Model;
using LMKit.TextGeneration;
using LMKit.TextGeneration.Chat;

using var model = LM.LoadFromModelID("qwen3:8b");
var conversation = new MultiTurnConversation(model)
{
ReasoningLevel = ReasoningLevel.High
};


var result = conversation.Submit("Solve: if 3x + 7 = 22, what is x?");
Console.WriteLine(result.Completion);

Remarks

When set to None, LM-Kit should omit any explicit “reasoning effort” hint. Use this when you want the model to decide its own internal effort or when targeting models that don’t expose a reasoning knob.

Typical interpretations:

  • Low - favor minimal internal deliberation for faster, cheaper replies.
  • Medium - balanced speed and quality (recommended default).
  • High - encourage more thorough internal reasoning; higher latency/cost is acceptable.
Share