Thinking Effort

Enhance your AI's reasoning with advanced thinking effort configurations.

Many reasoning models allow you to adjust the thinking effort of the model. This feature allows you to control how much time and effort the model spends reasoning about a response before generating it.

Please note that not all models support reasoning and that model providers may have different configurations for thinking effort. The "thinking" text may also not be returned for all models as well.

Set Model Purpose with Reasoning

In order to adjust reasoning and thinking effort, you must first enable Thinking as a model's purpose. This is done by editing a model from the model selection dropdown list, then selecting the edit icon, and then enabling Thinking.

model purpose

Set Thinking Effort

Once Thinking is enabled, you can set the thinking effort for the model. This can be done by selecting the Thinking Effort dropdown in the model settings.

thinking effort parameter

Providers

Providers may support different thinking effort configurations. The following providers are known to support thinking effort. However, not all models within these providers support thinking effort configurations. Please check the provider and their model documentation for more details. You will likely see an error message returned when chatting if you try to set thinking effort on a model that does not support it.

Key

Use the following key to understand the thinking effort support for each provider below.

  • None: no thinking effort is applied, model will respond immediately without any reasoning
  • Default: the default thinking effort configuration determined by the provider
  • Think Lightly: a light reasoning effort, model will take a short time to respond with minimal reasoning
  • Think Moderately: a moderate reasoning effort, model will take a longer time to respond with some reasoning
  • Think Deeply: a deep reasoning effort, model will take the longest time to respond with extensive reasoning
  • Dynamic: the model will dynamically adjust the thinking effort based on the complexity
  • Exclude: the model performs its default reasoning effort, but does not return the reasoning text

Ollama / Msty Models

Supports:

  • None
  • Default

OpenAI Compatible Models

Supports:

  • None
  • Default

OpenAI / ChatGPT

Supports:

  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply

Anthropic / Claude

Supports:

  • None
  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply

Google Generative AI / Gemini

Supports:

  • None
  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply
  • Dynamic

xAI

Supports:

  • None
  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply

OpenRouter

Supports:

  • None
  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply
  • Exclude

Mistral AI

Supports:

  • None
  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply

Perplexity

Supports:

  • None
  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply

Groq

Supports:

  • None
  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply

Cohere

Supports:

  • None
  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply

Azure OpenAI

Supports:

  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply

DeepSeek

Supports:

  • None
  • Default
  • Think Lightly
  • Think Moderately
  • Think Deeply