If you are planning to use Google's Gemini 3.5 model or its API, there are some changes you must be aware of! Instead of the thinking_budget parameter, which previously used integer values to define the depth of the thinking process in version 2.5, you must now use the more intuitive string option thinking_level.

Be careful, as using old workarounds that send both integer and string values, or sticking to the old integer configuration, will immediately result in a 400 error when calling the API. The new Gemini 3.5 Flash defaults to 'medium' for its thinking strength, which strikes a good balance between speed and token usage. For more complex coding or math problems, switching to 'high' can be quite effective.

Also, another pro tip to remember is that when you enable reasoning settings, you should keep the temperature at the default 1.0 if possible. Many people lower the temperature to get more consistent answers, but with reasoning models, lowering it can cause the model to get stuck in a loop, produce strange outputs, or see a significant drop in performance.

When writing code in a Python environment, you can set the new reasoning level option as shown in the example below for a clean execution.

# google-genai SDK v2.0.0 이상 기준
config = types.GenerateContentConfig(
    thinking_config=types.ThinkingConfig(thinking_level="high"),
    temperature=1.0
)

In the Google AI Studio interface, you can now select from a simple dropdown menu instead of dealing with complex sliders, so feel free to adjust the depth to suit your development projects or everyday reasoning tasks!