Thinking

Thinking models are trained to generate an internal "thinking process" before producing a response. As a result, thinking models are capable of stronger reasoning, multi-step planning, mathematical problem-solving, and code generation than models without thinking capabilities.

The thinking process is enabled by default across Gemini models. When you use Agent Studio on Gemini Enterprise Agent Platform, you can view the full thinking process together with the model's generated response.

Supported models

Thinking is supported in the following models:

Click to expand supported models

• Gemini Omni Flash preview
• Gemini 3.7 Flash
• Gemini 3.6 Flash
• Gemini 3.5 Flash-Lite
• Gemini 3.5 Flash
• Gemini 3.1 Pro preview
• Gemini 3.1 Flash-Lite Image (Nano Banana 2 Lite)
• Gemini 3.1 Flash-Lite
• Gemini 3.1 Flash Image
• Gemini 3 Pro Image
• Gemini 3 Flash preview
• Gemini 2.5 Pro
• Gemini 2.5 Flash-Lite
• Gemini 2.5 Flash

Control model thinking

You can control the amount of thinking the model performs before returning a response. The method for controlling thinking differs depending on the model version.

Gemini 3 and later models

Gemini 3 models introduce the thinking_level parameter, which simplifies thinking budget configuration into discrete levels. By default, Gemini 3 models use dynamic thinking (thinking_level.HIGH) to reason through prompts. For faster, lower-latency responses when complex reasoning isn't required, you can constrain the model's thinking_level.

The following table summarizes which thinking_level values are supported by each model, and the default thinking_level for each model:

Model	Supported thinking_level values	Default
Gemini 3.7 Flash	LOW, MEDIUM, HIGH	MEDIUM
Gemini 3.6 Flash	MINIMAL, LOW, MEDIUM, HIGH	MEDIUM
Gemini 3.5 Flash-Lite	MINIMAL, LOW, MEDIUM, HIGH	MINIMAL
Gemini 3.5 Flash	MINIMAL, LOW, MEDIUM, HIGH	MEDIUM
Gemini 3.1 Pro preview	LOW, MEDIUM, HIGH	HIGH
Gemini 3.1 Flash-Lite Image (Nano Banana 2 Lite)	MINIMAL, HIGH	MINIMAL
Gemini 3.1 Flash-Lite	MINIMAL, LOW, MEDIUM, HIGH	MINIMAL
Gemini 3.1 Flash Image	MINIMAL, HIGH	MINIMAL
Gemini 3 Pro Image	HIGH	HIGH
Gemini 3 Flash preview	MINIMAL, LOW, MEDIUM, HIGH	HIGH

• MINIMAL: Constrains the model to use as few tokens as possible for thinking and is best used for low-complexity tasks that wouldn't benefit from extensive reasoning. This is the default level for Gemini 3.1 Flash-Lite. MINIMAL is as close as possible to a zero budget for thinking but still requires thought signatures. If thought signatures aren't provided in your request, the model returns a 400 error. For more information, see Thought signatures.

  from google import genai
  from google.genai import types
  
  client = genai.Client()
  
  response = client.models.generate_content(
      model="gemini-3-flash-preview",
      contents="How does AI work?",
      config=types.GenerateContentConfig(
          thinking_config=types.ThinkingConfig(
              thinking_level=types.ThinkingLevel.MINIMAL
          )
      ),
  )
  print(response.text)
  

• LOW: Constrains the model to use fewer tokens for thinking and is suitable for simpler tasks where extensive reasoning is not required. LOW is ideal for high-throughput tasks where speed is essential:

  from google import genai
  from google.genai import types
  
  client = genai.Client()
  
  response = client.models.generate_content(
      model="gemini-3.5-flash",
      contents="How does AI work?",
      config=types.GenerateContentConfig(
          thinking_config=types.ThinkingConfig(
              thinking_level=types.ThinkingLevel.LOW
          )
      ),
  )
  print(response.text)
  

• MEDIUM: Offers a balanced approach suitable for tasks of moderate complexity that benefit from reasoning but don't require deep, multi-step planning. It provides more reasoning capability than LOW while maintaining lower latency than HIGH:

  from google import genai
  from google.genai import types
  
  client = genai.Client()
  
  response = client.models.generate_content(
      model="gemini-3-flash-preview",
      contents="How does AI work?",
      config=types.GenerateContentConfig(
          thinking_config=types.ThinkingConfig(
              thinking_level=types.ThinkingLevel.MEDIUM
          )
      ),
  )
  print(response.text)
  

• HIGH: Allows the model to use more tokens for thinking and is suitable for complex prompts requiring deep reasoning, such as multi-step planning, verified code generation, or advanced function calling scenarios. This is the default level for Gemini 3 Pro models and Gemini 3 Flash. Use this configuration when replacing tasks you might have previously relied on specialized reasoning models for:

  from google import genai
  from google.genai import types
  
  client = genai.Client()
  
  response = client.models.generate_content(
      model="gemini-3.5-flash",
      contents="Find the race condition in this multi-threaded C++ snippet: [code here]",
      config=types.GenerateContentConfig(
          thinking_config=types.ThinkingConfig(
              thinking_level=types.ThinkingLevel.HIGH
          )
      ),
  )
  print(response.text)
  

Thinking cannot be turned off for Gemini 3 Pro and Gemini 3.1 Pro.

If you specify both thinking_level and thinking_budget in the same request for a Gemini 3 model, the model returns an error.

Gemini 2.5 and earlier models

For models earlier than Gemini 3, you can control thinking using the thinking_budget parameter, which sets an upper limit on the number of tokens the model can use for its thought process. By default, if thinking_budget is not set, the model automatically controls how much it thinks up to a maximum of 8,192 tokens. To use dynamic budget through the API, set thinking_budget to -1.

You can manually set thinking_budget to impose a soft upper limit on the number of tokens in situations where you might need more or less tokens than the default thinking budget. You can set a lower token limit for less complex tasks, or a higher limit for more complex ones. Note that this is a soft limit and therefore there can be variability in total thought tokens. If latency is more important, use a lower budget or set the budget to 0 to prevent thought content from being returned with the response.

The following table shows the minimum and maximum amounts you can set the thinking_budget to for each supported model, and the default thinking budget for each model:

Model	Minimum token amount	Maximum token amount	Default
Gemini 2.5 Flash	1	24,576	Auto (up to 8,192 tokens)
Gemini 2.5 Pro	128	32,768	Auto (up to 8,192 tokens)
Gemini 2.5 Flash-Lite	512	24,576	Auto (up to 8,192 tokens)

If you set thinking_budget to 0 when using Gemini 2.5 Flash and Gemini 2.5 Flash-Lite, no thought content is returned with the response. However, reasoning-style text might still be present in the model's output. Thinking can't be turned off for Gemini 2.5 Pro.

If you use the thinking_level parameter with a model earlier than Gemini 3, the model returns an error.

Console

1. Open Agent Studio > Create prompt.
2. In the Model panel, click Switch model and select one of the supported models from the menu.
3. Select Manual from the Thinking budget drop-down selector and then use the slider to adjust the thinking budget limit.

Python

Install

pip install --upgrade google-genai

To learn more, see the SDK reference documentation.

Set environment variables to use the Google Gen AI SDK with Vertex AI:

# Replace the `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION` values
# with appropriate values for your project.
export GOOGLE_CLOUD_PROJECT=GOOGLE_CLOUD_PROJECT
export GOOGLE_CLOUD_LOCATION=global
export GOOGLE_GENAI_USE_ENTERPRISE=True

from google import genai
from google.genai.types import GenerateContentConfig, ThinkingConfig

client = genai.Client()

response = client.models.generate_content(
    model="gemini-3.5-flash",
    contents="solve x^2 + 4x + 4 = 0",
    config=GenerateContentConfig(
        thinking_config=ThinkingConfig(
            thinking_budget=1024,  # Use `0` to turn off thinking
        )
    ),
)

print(response.text)
# Example response:
#     To solve the equation $x^2 + 4x + 4 = 0$, you can use several methods:
#     **Method 1: Factoring**
#     1.  Look for two numbers that multiply to the constant term (4) and add up to the coefficient of the $x$ term (4).
#     2.  The numbers are 2 and 2 ($2 \times 2 = 4$ and $2 + 2 = 4$).
#     ...
#     ...
#     All three methods yield the same solution. This quadratic equation has exactly one distinct solution (a repeated root).
#     The solution is **x = -2**.

# Token count for `Thinking`
print(response.usage_metadata.thoughts_token_count)
# Example response:
#     886

# Total token count
print(response.usage_metadata.total_token_count)
# Example response:
#     1525

Node.js

Install

npm install @google/genai

To learn more, see the SDK reference documentation.

Set environment variables to use the Google Gen AI SDK with Vertex AI:

# Replace the `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION` values
# with appropriate values for your project.
export GOOGLE_CLOUD_PROJECT=GOOGLE_CLOUD_PROJECT
export GOOGLE_CLOUD_LOCATION=global
export GOOGLE_GENAI_USE_ENTERPRISE=True

const {GoogleGenAI} = require('@google/genai');

const GOOGLE_CLOUD_PROJECT = process.env.GOOGLE_CLOUD_PROJECT;
const GOOGLE_CLOUD_LOCATION = process.env.GOOGLE_CLOUD_LOCATION || 'global';

async function generateWithThoughts(
  projectId = GOOGLE_CLOUD_PROJECT,
  location = GOOGLE_CLOUD_LOCATION
) {
  const client = new GoogleGenAI({
    vertexai: true,
    project: projectId,
    location: location,
  });

  const response = await client.models.generateContent({
    model: 'gemini-2.5-flash',
    contents: 'solve x^2 + 4x + 4 = 0',
    config: {
      thinkingConfig: {
        thinkingBudget: 1024,
      },
    },
  });

  console.log(response.text);
  // Example response:
  //  To solve the equation $x^2 + 4x + 4 = 0$, you can use several methods:
  //  **Method 1: Factoring**
  //  1.  Look for two numbers that multiply to the constant term (4) and add up to the coefficient of the $x$ term (4).
  //  2.  The numbers are 2 and 2 ($2 \times 2 = 4$ and $2 + 2 = 4$).
  //  ...
  //  ...
  //  All three methods yield the same solution. This quadratic equation has exactly one distinct solution (a repeated root).
  //  The solution is **x = -2**.

  // Token count for `Thinking`
  console.log(response.usageMetadata.thoughtsTokenCount);
  // Example response:
  //  886

  // Total token count
  console.log(response.usageMetadata.totalTokenCount);
  // Example response:
  //  1525
  return response.text;
}

Go

Learn how to install or update the Go.

To learn more, see the SDK reference documentation.

Set environment variables to use the Google Gen AI SDK with Vertex AI:

# Replace the `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION` values
# with appropriate values for your project.
export GOOGLE_CLOUD_PROJECT=GOOGLE_CLOUD_PROJECT
export GOOGLE_CLOUD_LOCATION=global
export GOOGLE_GENAI_USE_ENTERPRISE=True

import (
	"context"
	"fmt"
	"io"

	"google.golang.org/genai"
)

// generateThinkingBudgetContentWithText demonstrates how to generate text including the model's thought process.
func generateThinkingBudgetContentWithText(w io.Writer) error {
	ctx := context.Background()

	client, err := genai.NewClient(ctx, &genai.ClientConfig{
		HTTPOptions: genai.HTTPOptions{APIVersion: "v1"},
	})
	if err != nil {
		return fmt.Errorf("failed to create genai client: %w", err)
	}

	modelName := "gemini-2.5-flash"
	thinkingBudget := int32(1024) //Use `0` to turn off thinking
	contents := []*genai.Content{
		{
			Parts: []*genai.Part{
				{Text: "solve x^2 + 4x + 4 = 0"},
			},
			Role: "user",
		},
	}

	resp, err := client.Models.GenerateContent(ctx,
		modelName,
		contents,
		&genai.GenerateContentConfig{
			ThinkingConfig: &genai.ThinkingConfig{
				ThinkingBudget: &thinkingBudget,
			},
		},
	)
	if err != nil {
		return fmt.Errorf("generate content failed: %w", err)
	}

	if resp.UsageMetadata != nil {
		fmt.Fprintf(w, "Thoughts token count: %d\n", resp.UsageMetadata.ThoughtsTokenCount)
		//Example response:
		//  908
		fmt.Fprintf(w, "Total token count: %d\n", resp.UsageMetadata.TotalTokenCount)
		//Example response:
		//  1364
	}

	fmt.Fprintln(w, resp.Text())

	// Example response:
	//    To solve the equation $x^2 + 4x + 4 = 0$, you can use several methods:
	//    **Method 1: Factoring**
	//    1.  Look for two numbers that multiply to the constant term (4) and add up to the coefficient of the $x$ term (4).
	//    2.  The numbers are 2 and 2 ($2 \times 2 = 4$ and $2 + 2 = 4$).
	//    ...
	//    ...
	//    Both methods yield the same result.
	//    The solution to the equation $x^2 + 4x + 4 = 0$ is **$x = -2$**.

	return nil
}

Java

Learn how to install or update the Java.

To learn more, see the SDK reference documentation.

Set environment variables to use the Google Gen AI SDK with Vertex AI:

# Replace the `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION` values
# with appropriate values for your project.
export GOOGLE_CLOUD_PROJECT=GOOGLE_CLOUD_PROJECT
export GOOGLE_CLOUD_LOCATION=global
export GOOGLE_GENAI_USE_ENTERPRISE=True

import com.google.genai.Client;
import com.google.genai.types.GenerateContentConfig;
import com.google.genai.types.GenerateContentResponse;
import com.google.genai.types.HttpOptions;
import com.google.genai.types.ThinkingConfig;

public class ThinkingBudgetWithTxt {

  public static void main(String[] args) {
    // TODO(developer): Replace these variables before running the sample.
    String modelId = "gemini-2.5-flash";
    generateContent(modelId);
  }

  // Generates text controlling the thinking budget
  public static String generateContent(String modelId) {
    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests.
    try (Client client =
        Client.builder()
            .location("global")
            .vertexAI(true)
            .httpOptions(HttpOptions.builder().apiVersion("v1").build())
            .build()) {

      GenerateContentConfig contentConfig =
          GenerateContentConfig.builder()
              .thinkingConfig(ThinkingConfig.builder().thinkingBudget(1024).build())
              .build();

      GenerateContentResponse response =
          client.models.generateContent(modelId, "solve x^2 + 4x + 4 = 0", contentConfig);

      System.out.println(response.text());
      // Example response:
      // To solve the equation $x^2 + 4x + 4 = 0$, we can use several methods:
      //
      // **Method 1: Factoring (Recognizing a Perfect Square Trinomial)**
      //
      // Notice that the left side of the equation is a perfect square trinomial. It fits the form
      // $a^2 + 2ab + b^2 = (a+b)^2$...
      // ...
      // The solution is $x = -2$.

      response
          .usageMetadata()
          .ifPresent(
              metadata -> {
                System.out.println("Token count for thinking: " + metadata.thoughtsTokenCount());
                System.out.println("Total token count: " + metadata.totalTokenCount());
              });
      // Example response:
      // Token count for thinking: Optional[885]
      // Total token count: Optional[1468]
      return response.text();
    }
  }
}

View thought summaries

Thought summaries provide visibility into the intermediate reasoning steps the model performed while generating a response. You can view thought summaries in Gemini 2.5 and newer models.

In Agent Studio, thought summaries are enabled by default and viewable by expanding the Thoughts panel.

When using the API, you can enable thought summaries by setting include_thoughts=True in your ThinkingConfig:

Console

Thought summaries are enabled by default in Agent Studio. You can see the model's summarized thought process by expanding the Thoughts panel.

Python

Install

pip install --upgrade google-genai

To learn more, see the SDK reference documentation.

Set environment variables to use the Google Gen AI SDK with Vertex AI:

# Replace the `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION` values
# with appropriate values for your project.
export GOOGLE_CLOUD_PROJECT=GOOGLE_CLOUD_PROJECT
export GOOGLE_CLOUD_LOCATION=global
export GOOGLE_GENAI_USE_ENTERPRISE=True

from google import genai
from google.genai.types import GenerateContentConfig, ThinkingConfig

client = genai.Client()
response = client.models.generate_content(
    model="gemini-3.1-pro-preview",
    contents="solve x^2 + 4x + 4 = 0",
    config=GenerateContentConfig(
        thinking_config=ThinkingConfig(include_thoughts=True)
    ),
)

print(response.text)
# Example Response:
#     Okay, let's solve the quadratic equation x² + 4x + 4 = 0.
#     ...
#     **Answer:**
#     The solution to the equation x² + 4x + 4 = 0 is x = -2. This is a repeated root (or a root with multiplicity 2).

for part in response.candidates[0].content.parts:
    if part and part.thought:  # show thoughts
        print(part.text)
# Example Response:
#     **My Thought Process for Solving the Quadratic Equation**
#
#     Alright, let's break down this quadratic, x² + 4x + 4 = 0. First things first:
#     it's a quadratic; the x² term gives it away, and we know the general form is
#     ax² + bx + c = 0.
#
#     So, let's identify the coefficients: a = 1, b = 4, and c = 4. Now, what's the
#     most efficient path to the solution? My gut tells me to try factoring; it's
#     often the fastest route if it works. If that fails, I'll default to the quadratic
#     formula, which is foolproof. Completing the square? It's good for deriving the
#     formula or when factoring is difficult, but not usually my first choice for
#     direct solving, but it can't hurt to keep it as an option.
#
#     Factoring, then. I need to find two numbers that multiply to 'c' (4) and add
#     up to 'b' (4). Let's see... 1 and 4 don't work (add up to 5). 2 and 2? Bingo!
#     They multiply to 4 and add up to 4. This means I can rewrite the equation as
#     (x + 2)(x + 2) = 0, or more concisely, (x + 2)² = 0. Solving for x is now
#     trivial: x + 2 = 0, thus x = -2.
#
#     Okay, just to be absolutely certain, I'll run the quadratic formula just to
#     double-check. x = [-b ± √(b² - 4ac)] / 2a. Plugging in the values, x = [-4 ±
#     √(4² - 4 * 1 * 4)] / (2 * 1). That simplifies to x = [-4 ± √0] / 2. So, x =
#     -2 again – a repeated root. Nice.
#
#     Now, let's check via completing the square. Starting from the same equation,
#     (x² + 4x) = -4. Take half of the b-value (4/2 = 2), square it (2² = 4), and
#     add it to both sides, so x² + 4x + 4 = -4 + 4. Which simplifies into (x + 2)²
#     = 0. The square root on both sides gives us x + 2 = 0, therefore x = -2, as
#      expected.
#
#     Always, *always* confirm! Let's substitute x = -2 back into the original
#     equation: (-2)² + 4(-2) + 4 = 0. That's 4 - 8 + 4 = 0. It checks out.
#
#     Conclusion: the solution is x = -2. Confirmed.

Node.js

Install

npm install @google/genai

To learn more, see the SDK reference documentation.

Set environment variables to use the Google Gen AI SDK with Vertex AI:

# Replace the `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION` values
# with appropriate values for your project.
export GOOGLE_CLOUD_PROJECT=GOOGLE_CLOUD_PROJECT
export GOOGLE_CLOUD_LOCATION=global
export GOOGLE_GENAI_USE_ENTERPRISE=True

const {GoogleGenAI} = require('@google/genai');

const GOOGLE_CLOUD_PROJECT = process.env.GOOGLE_CLOUD_PROJECT;
const GOOGLE_CLOUD_LOCATION = process.env.GOOGLE_CLOUD_LOCATION || 'global';

async function generateWithThoughts(
  projectId = GOOGLE_CLOUD_PROJECT,
  location = GOOGLE_CLOUD_LOCATION
) {
  const client = new GoogleGenAI({
    vertexai: true,
    project: projectId,
    location: location,
  });

  const response = await client.models.generateContent({
    model: 'gemini-2.5-pro',
    contents: 'solve x^2 + 4x + 4 = 0',
    config: {
      thinkingConfig: {
        includeThoughts: true,
      },
    },
  });

  console.log(response.text);
  // Example Response:
  //  Okay, let's solve the quadratic equation x² + 4x + 4 = 0.
  //  ...
  //  **Answer:**
  //  The solution to the equation x² + 4x + 4 = 0 is x = -2. This is a repeated root (or a root with multiplicity 2).

  for (const part of response.candidates[0].content.parts) {
    if (part && part.thought) {
      console.log(part.text);
    }
  }

  // Example Response:
  // **My Thought Process for Solving the Quadratic Equation**
  //
  // Alright, let's break down this quadratic, x² + 4x + 4 = 0. First things first:
  // it's a quadratic; the x² term gives it away, and we know the general form is
  // ax² + bx + c = 0.
  //
  // So, let's identify the coefficients: a = 1, b = 4, and c = 4. Now, what's the
  // most efficient path to the solution? My gut tells me to try factoring; it's
  // often the fastest route if it works. If that fails, I'll default to the quadratic
  // formula, which is foolproof. Completing the square? It's good for deriving the
  // formula or when factoring is difficult, but not usually my first choice for
  // direct solving, but it can't hurt to keep it as an option.
  //
  // Factoring, then. I need to find two numbers that multiply to 'c' (4) and add
  // up to 'b' (4). Let's see... 1 and 4 don't work (add up to 5). 2 and 2? Bingo!
  // They multiply to 4 and add up to 4. This means I can rewrite the equation as
  // (x + 2)(x + 2) = 0, or more concisely, (x + 2)² = 0. Solving for x is now
  // trivial: x + 2 = 0, thus x = -2.
  //
  // Okay, just to be absolutely certain, I'll run the quadratic formula just to
  // double-check. x = [-b ± √(b² - 4ac)] / 2a. Plugging in the values, x = [-4 ±
  // √(4² - 4 * 1 * 4)] / (2 * 1). That simplifies to x = [-4 ± √0] / 2. So, x =
  // -2 again – a repeated root. Nice.
  //
  // Now, let's check via completing the square. Starting from the same equation,
  // (x² + 4x) = -4. Take half of the b-value (4/2 = 2), square it (2² = 4), and
  // add it to both sides, so x² + 4x + 4 = -4 + 4. Which simplifies into (x + 2)²
  // = 0. The square root on both sides gives us x + 2 = 0, therefore x = -2, as
  //  expected.
  //
  // Always, *always* confirm! Let's substitute x = -2 back into the original
  // equation: (-2)² + 4(-2) + 4 = 0. That's 4 - 8 + 4 = 0. It checks out.
  //
  // Conclusion: the solution is x = -2. Confirmed.

  return response.text;
}

Go

Learn how to install or update the Go.

To learn more, see the SDK reference documentation.

Set environment variables to use the Google Gen AI SDK with Vertex AI:

# Replace the `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION` values
# with appropriate values for your project.
export GOOGLE_CLOUD_PROJECT=GOOGLE_CLOUD_PROJECT
export GOOGLE_CLOUD_LOCATION=global
export GOOGLE_GENAI_USE_ENTERPRISE=True

import (
	"context"
	"fmt"
	"io"

	"google.golang.org/genai"
)

// generateContentWithThoughts demonstrates how to generate text including the model's thought process.
func generateContentWithThoughts(w io.Writer) error {
	ctx := context.Background()

	client, err := genai.NewClient(ctx, &genai.ClientConfig{
		HTTPOptions: genai.HTTPOptions{APIVersion: "v1"},
	})
	if err != nil {
		return fmt.Errorf("failed to create genai client: %w", err)
	}

	modelName := "gemini-2.5-pro"
	contents := []*genai.Content{
		{
			Parts: []*genai.Part{
				{Text: "solve x^2 + 4x + 4 = 0"},
			},
			Role: "user",
		},
	}

	resp, err := client.Models.GenerateContent(ctx,
		modelName,
		contents,
		&genai.GenerateContentConfig{
			ThinkingConfig: &genai.ThinkingConfig{
				IncludeThoughts: true,
			},
		},
	)
	if err != nil {
		return fmt.Errorf("failed to generate content: %w", err)
	}

	if len(resp.Candidates) == 0 || resp.Candidates[0].Content == nil {
		return fmt.Errorf("no content was generated")
	}

	// The response may contain both the final answer and the model's thoughts.
	// Iterate through the parts to print them separately.
	fmt.Fprintln(w, "Answer:")
	for _, part := range resp.Candidates[0].Content.Parts {
		if part.Text != "" && !part.Thought {
			fmt.Fprintln(w, part.Text)
		}
	}
	fmt.Fprintln(w, "\nThoughts:")
	for _, part := range resp.Candidates[0].Content.Parts {
		if part.Thought {
			fmt.Fprintln(w, part.Text)
		}
	}

	// Example response:
	//  Answer:
	//	Of course! We can solve this quadratic equation in a couple of ways.
	//
	//### Method 1: Factoring (the easiest method for this problem)
	//
	//1.  **Recognize the pattern.** The expression `x² + 4x + 4` is a perfect square trinomial. It fits the pattern `a² + 2ab + b² = (a + b)²`. In this case, `a = x` and `b = 2`.
	//
	//2.  **Factor the equation.**
	//    `x² + 4x + 4 = (x + 2)(x + 2) = (x + 2)²`
	//
	//3.  **Solve for x.** Now set the factored expression to zero:
	//    `(x + 2)² = 0`
	//
	//    Take the square root of both sides:
	//    `x + 2 = 0`
	//
	//    Subtract 2 from both sides:
	//    `x = -2`
	//
	//This type of solution is called a "repeated root" or a "double root" because the factor `(x+2)` appears twice.
	//
	//---
	//
	//### Method 2: Using the Quadratic Formula
	//
	//You can use the quadratic formula for any equation in the form `ax² + bx + c = 0`.
	//
	//The formula is: `x = [-b ± sqrt(b² - 4ac)] / 2a`
	//
	//1.  **Identify a, b, and c.**
	//    *   a = 1
	//    *   b = 4
	//    *   c = 4
	//
	//2.  **Plug the values into the formula.**
	//    `x = [-4 ± sqrt(4² - 4 * 1 * 4)] / (2 * 1)`
	//
	//3.  **Simplify.**
	//    `x = [-4 ± sqrt(16 - 16)] / 2`
	//    `x = [-4 ± sqrt(0)] / 2`
	//    `x = -4 / 2`
	//
	//4.  **Solve for x.**
	//    `x = -2`
	//Alright, the user wants to solve the quadratic equation `x² + 4x + 4 = 0`. My first instinct is to see if I can factor it; that's often the fastest approach if it works.  Looking at the coefficients, I see `a = 1`, `b = 4`, and `c = 4`.  Factoring is clearly the most direct path here. I need to find two numbers that multiply to 4 (c) and add up to 4 (b). Hmm, let's see… 1 and 4? Nope, that adds to 5.  2 and 2? Perfect!  2 times 2 is 4, and 2 plus 2 is also 4.
	//
	//So, `x² + 4x + 4` factors nicely into `(x + 2)(x + 2)`.  Ah, a perfect square trinomial! That's useful to note. Now, I can write the equation as `(x + 2)² = 0`.  Taking the square root of both sides gives me `x + 2 = 0`.  And finally, subtracting 2 from both sides, I get `x = -2`.  That's the solution.
	//
	//Just to be thorough, and maybe to offer an alternative explanation, let's verify this using the quadratic formula. It's `x = [-b ± √(b² - 4ac)] / 2a`. Plugging in my values:  `x = [-4 ± √(4² - 4 * 1 * 4)] / (2 * 1)`.  That simplifies to `x = [-4 ± √(16 - 16)] / 2`, or `x = [-4 ± 0] / 2`.  Therefore, `x = -2`. The discriminant being zero tells me I have exactly one real, repeated root.  Great. So, whether I factor or use the quadratic formula, the answer is the same.
	return nil
}

Java

Learn how to install or update the Java.

To learn more, see the SDK reference documentation.

Set environment variables to use the Google Gen AI SDK with Vertex AI:

# Replace the `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION` values
# with appropriate values for your project.
export GOOGLE_CLOUD_PROJECT=GOOGLE_CLOUD_PROJECT
export GOOGLE_CLOUD_LOCATION=global
export GOOGLE_GENAI_USE_ENTERPRISE=True

import com.google.genai.Client;
import com.google.genai.types.Candidate;
import com.google.genai.types.Content;
import com.google.genai.types.GenerateContentConfig;
import com.google.genai.types.GenerateContentResponse;
import com.google.genai.types.HttpOptions;
import com.google.genai.types.ThinkingConfig;

public class ThinkingIncludeThoughtsWithTxt {

  public static void main(String[] args) {
    // TODO(developer): Replace these variables before running the sample.
    String modelId = "gemini-2.5-pro";
    generateContent(modelId);
  }

  // Generates text including thoughts in the response
  public static String generateContent(String modelId) {
    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests.
    try (Client client =
        Client.builder()
            .location("global")
            .vertexAI(true)
            .httpOptions(HttpOptions.builder().apiVersion("v1").build())
            .build()) {

      GenerateContentConfig contentConfig =
          GenerateContentConfig.builder()
              .thinkingConfig(ThinkingConfig.builder().includeThoughts(true).build())
              .build();

      GenerateContentResponse response =
          client.models.generateContent(modelId, "solve x^2 + 4x + 4 = 0", contentConfig);

      System.out.println(response.text());
      // Example response:
      // We can solve the equation x² + 4x + 4 = 0 using a couple of common methods.
      //
      // ### Method 1: Factoring (The Easiest Method for this Problem)
      // **Recognize the pattern:** The pattern for a perfect square trinomial
      // is a² + 2ab + b² = (a + b)².
      // ...
      // ### Final Answer:
      // The solution is **x = -2**.

      // Get parts of the response and print thoughts
      response
          .candidates()
          .flatMap(candidates -> candidates.stream().findFirst())
          .flatMap(Candidate::content)
          .flatMap(Content::parts)
          .ifPresent(
              parts -> {
                parts.forEach(
                    part -> {
                      if (part.thought().orElse(false)) {
                        part.text().ifPresent(System.out::println);
                      }
                    });
              });
      // Example response:
      // Alright, let's break down this quadratic equation, x² + 4x + 4 = 0. My initial thought is,
      // "classic quadratic."  I'll need to find the values of 'x' that make this equation true. The
      // equation is in standard form, and since the coefficients are relatively small, I
      // immediately suspect that factoring might be the easiest route.  It's worth checking.
      //
      // First, I assessed what I had. *a* is 1, *b* is 4, and *c* is 4. I consider my toolkit.
      // Factoring is the likely first choice, then I can use the quadratic formula as a backup,
      // because that ALWAYS works, and I could use graphing. However, for this, factoring seems the
      // cleanest approach.
      //
      // Okay, factoring. I need two numbers that multiply to *c* (which is 4) and add up to *b*
      // (also 4).  I quickly run through the factor pairs of 4: (1, 4), (-1, -4), (2, 2), (-2, -2).
      //  Aha! 2 and 2 fit the bill. They multiply to 4 *and* add up to 4.  Therefore, I can rewrite
      // the equation as (x + 2)(x + 2) = 0.  That simplifies to (x + 2)² = 0. Perfect square
      // trinomial – nice and tidy. Seeing that pattern from the outset can save a step or two. Now,
      // to solve for *x*:  if (x + 2)² = 0, then x + 2 must equal 0.  Therefore, x = -2. Done.
      //
      // But, for the sake of a full explanation, let's use the quadratic formula as a second
      // method. It's a reliable way to double-check the answer, plus it's good practice.  I plug my
      // *a*, *b*, and *c* values into the formula: x = [-b ± √(b² - 4ac)] / (2a). That gives me  x
      // = [-4 ± √(4² - 4 * 1 * 4)] / (2 * 1). Simplifying under the radical, I get x = [-4 ± √(16 -
      // 16)] / 2. So, x = [-4 ± √0] / 2. The square root of 0 is zero, which is very telling!  When
      // the discriminant (b² - 4ac) is zero, you get one real solution, a repeated root. This means
      // x = -4 / 2, which simplifies to x = -2.  Exactly the same as before.
      //
      // Therefore, the answer is x = -2.  Factoring was the most straightforward route.  For
      // completeness, I showed the solution via the quadratic formula, too. Both approaches lead to
      // the same single solution.  This is a repeated root – a double root, if you will.
      //
      // And to be absolutely sure...let's check our answer! Substitute -2 back into the original
      // equation. (-2)² + 4(-2) + 4 = 4 - 8 + 4 = 0.  Yep, 0 = 0. The solution is correct.
      return response.text();
    }
  }
}

A response may contain a thought signature without thought summary text in the following scenarios:

• Low-complexity requests: The model required minimal reasoning steps to formulate the response.
• Disabled summaries: Thought summaries were not requested or were explicitly turned off.
• Non-text reasoning modalities: Certain modalities (such as image processing) might not emit text summaries.

Your application should always gracefully handle responses where thought summary content is absent or empty while preserving the associated thought signatures.

Thought signatures

Thought signatures are encrypted representations of the model's internal thought process that preserve the Gemini reasoning state during multi-turn conversations, especially when using function calling.

To ensure the model maintains full context across multiple turns of a conversation, you must return the thought signatures from previous responses in your subsequent requests, regardless of the thinking level used. If you are using the official Google Google Gen AI SDK (Python, Node.js, Go, or Java) and using the standard chat history features or appending the full model response to the history, thought signatures are handled automatically.

For detailed rules, examples, and multi-turn workflow patterns, see Thought signatures.

Prompting techniques

Effective prompt design helps you steer model reasoning, reserve token budget, and achieve optimal output quality with thinking models.

For comprehensive strategies, multishot patterns, verification prompts, and debugging tips, see the Thinking prompting guide.

Pricing

You are charged for the tokens that are generated during a model's thinking process. For some models, such as Gemini 3 Pro and Gemini 2.5 Pro, thinking is enabled by default and you are billed for these tokens.

For more information, see Agent Platform pricing. To learn how to manage costs, see Control model thinking.

What's next

Guide

Thought signatures

Learn how to preserve the Gemini reasoning state during multi-turn and multi-step conversations using thought signatures.

Guide

Thinking prompting guide

Explore prompt engineering techniques and best practices tailored for Gemini thinking models.

Console

Google Cloud Console

Try prompting Gemini for yourself in the Google Cloud Console.