Problem

For reasoning models, the reasoning_effort parameter is currently not set. This means the model decides on its own how much “thinking effort” to apply.
In most cases, the model defaults to a medium level, which generally works fine. However, this can lead to higher token usage and is not always optimal for all use cases.
At the moment, users have no way to control this behavior, even though different scenarios may require different levels of reasoning (e.g. lower for simple translations, higher for complex tasks).

Solution

Add a configuration option for reasoning_effort in the model settings. Allow users to choose the desired reasoning level (e.g. none, low, medium, high), depending on what the model supports. Only show and enable this option for models that actually support the parameter. Use the configured value when sending requests to the model- This gives users more control over performance, cost, and output behavior.