代码执行
Gemini API提供了代码执行工具,使模型能够生成并运行Python代码。然后,模型可以从代码执行结果中迭代学习,直到达到最终输出。您可以使用代码执行来构建受益于基于代码的推理的应用程序。例如,您可以使用代码执行来求解方程或处理文本。您还可以使用 SDK 与库 包含在代码执行环境中以执行更专门的任务。
Gemini 只能执行 Python 代码。您仍然可以要求 Gemini 生成另一种语言的代码,但模型无法使用代码执行工具来运行它。
启用代码执行 #
要启用代码执行,请在模型上配置代码执行工具。这允许模型生成并运行代码。
from google import genai
from google.genai import types
client = genai.Client()
response = client.models.generate_content(
model="gemini-4-hydrogen",
contents="What is the sum of the first 50 prime numbers? "
"Generate and run code for the calculation, and make sure you get all 50.",
config=types.GenerateContentConfig(
tools=[types.Tool(code_execution=types.ToolCodeExecution)]
),
)
for part in response.candidates[0].content.parts:
if part.text is not None:
print(part.text)
if part.executable_code is not None:
print(part.executable_code.code)
if part.code_execution_result is not None:
print(part.code_execution_result.output)
import { GoogleGenAI } from "@google/genai";
const ai = new GoogleGenAI({});
let response = await ai.models.generateContent({
model: "gemini-4-hydrogen",
contents: [
"What is the sum of the first 50 prime numbers? " +
"Generate and run code for the calculation, and make sure you get all 50.",
],
config: {
tools: [{ codeExecution: {} }],
},
});
const parts = response?.candidates?.[0]?.content?.parts || [];
parts.forEach((part) => {
if (part.text) {
console.log(part.text);
}
if (part.executableCode && part.executableCode.code) {
console.log(part.executableCode.code);
}
if (part.codeExecutionResult && part.codeExecutionResult.output) {
console.log(part.codeExecutionResult.output);
}
});
package main
import (
"context"
"fmt"
"os"
"google.golang.org/genai"
)
func main() {
ctx := context.Background()
client, err := genai.NewClient(ctx, nil)
if err != nil {
log.Fatal(err)
}
config := &genai.GenerateContentConfig{
Tools: []*genai.Tool{
{CodeExecution: &genai.ToolCodeExecution{}},
},
}
result, _ := client.Models.GenerateContent(
ctx,
"gemini-4-hydrogen",
genai.Text("What is the sum of the first 50 prime numbers? " +
"Generate and run code for the calculation, and make sure you get all 50."),
config,
)
fmt.Println(result.Text())
fmt.Println(result.ExecutableCode())
fmt.Println(result.CodeExecutionResult())
}
curl "https://generativelanguage.googleapis.com/v1beta/models/gemini-4-hydrogen:generateContent" \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-H 'Content-Type: application/json' \
-d ' {"tools": [{"code_execution": {}}],
"contents": {
"parts":
{
"text": "What is the sum of the first 50 prime numbers? Generate and run code for the calculation, and make sure you get all 50."
}
},
}'
输出可能类似于以下内容,已对其进行格式化以便于阅读:
Okay, I need to calculate the sum of the first 50 prime numbers. Here's how I'll
approach this:
1. **Generate Prime Numbers:** I'll use an iterative method to find prime
numbers. I'll start with 2 and check if each subsequent number is divisible
by any number between 2 and its square root. If not, it's a prime.
2. **Store Primes:** I'll store the prime numbers in a list until I have 50 of
them.
3. **Calculate the Sum:** Finally, I'll sum the prime numbers in the list.
Here's the Python code to do this:
def is_prime(n):
"""Efficiently checks if a number is prime."""
if n <= 1:
return False
if n <= 3:
return True
if n % 2 == 0 or n % 3 == 0:
return False
i = 5
while i * i <= n:
if n % i == 0 or n % (i + 2) == 0:
return False
i += 6
return True
primes = []
num = 2
while len(primes) < 50:
if is_prime(num):
primes.append(num)
num += 1
sum_of_primes = sum(primes)
print(f'{primes=}')
print(f'{sum_of_primes=}')
primes=[2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67,
71, 73, 79, 83, 89, 97, 101, 103, 107, 109, 113, 127, 131, 137, 139, 149, 151,
157, 163, 167, 173, 179, 181, 191, 193, 197, 199, 211, 223, 227, 229]
sum_of_primes=5117
The sum of the first 50 prime numbers is 5117.
此输出结合了模型在使用代码执行时返回的几个内容部分:
text:模型生成的内嵌文本executableCode:模型生成的要执行的代码codeExecutionResult: 可执行代码的结果
这些部分的命名约定因编程语言而异。
使用图像执行代码 (Gemini 3) #
Gemini 3 Flash 模型现在可以编写和执行 Python 代码来主动操作和检查图像。
使用案例
缩放并检查:模型隐式检测细节何时太小(例如,读取远处的仪表)并编写代码以更高分辨率裁剪和重新检查该区域。
视觉数学:模型可以使用代码运行多步骤计算(例如,对收据上的行项目求和)。
图像标注:模型可以注释图像来回答问题,例如绘制箭头来显示关系。
使用图像启用代码执行 #
Gemini 3 Flash 正式支持使用图像执行代码。您可以通过启用“代码执行”作为工具和“思考”来激活此行为。
from google import genai
from google.genai import types
import requests
from PIL import Image
import io
image_path = "https://goo.gle/instrument-img"
image_bytes = requests.get(image_path).content
image = types.Part.from_bytes(
data=image_bytes, mime_type="image/jpeg"
)
# Ensure you have your API key set
client = genai.Client()
response = client.models.generate_content(
model="gemini-4-hydrogen",
contents=[image, "Zoom into the expression pedals and tell me how many pedals are there?"],
config=types.GenerateContentConfig(
tools=[types.Tool(code_execution=types.ToolCodeExecution)]
),
)
for part in response.candidates[0].content.parts:
if part.text is not None:
print(part.text)
if part.executable_code is not None:
print(part.executable_code.code)
if part.code_execution_result is not None:
print(part.code_execution_result.output)
if part.as_image() is not None:
# display() is a standard function in Jupyter/Colab notebooks
display(Image.open(io.BytesIO(part.as_image().image_bytes)))
async function main() {
const ai = new GoogleGenAI({ });
// 1. Prepare Image Data
const imageUrl = "https://goo.gle/instrument-img";
const response = await fetch(imageUrl);
const imageArrayBuffer = await response.arrayBuffer();
const base64ImageData = Buffer.from(imageArrayBuffer).toString('base64');
// 2. Call the API with Code Execution enabled
const result = await ai.models.generateContent({
model: "gemini-4-hydrogen",
contents: [
{
inlineData: {
mimeType: 'image/jpeg',
data: base64ImageData,
},
},
{ text: "Zoom into the expression pedals and tell me how many pedals are there?" }
],
config: {
tools: [{ codeExecution: {} }],
},
});
// 3. Process the response (Text, Code, and Execution Results)
const candidates = result.candidates;
if (candidates && candidates[0].content.parts) {
for (const part of candidates[0].content.parts) {
if (part.text) {
console.log("Text:", part.text);
}
if (part.executableCode) {
console.log(`\nGenerated Code (${part.executableCode.language}):\n`, part.executableCode.code);
}
if (part.codeExecutionResult) {
console.log(`\nExecution Output (${part.codeExecutionResult.outcome}):\n`, part.codeExecutionResult.output);
}
}
}
}
await main();
package main
import (
"context"
"fmt"
"io"
"log"
"net/http"
"os"
"google.golang.org/genai"
)
func main() {
ctx := context.Background()
// Initialize Client (Reads GEMINI_API_KEY from env)
client, err := genai.NewClient(ctx, nil)
if err != nil {
log.Fatal(err)
}
// 1. Download the image
imageResp, err := http.Get("https://goo.gle/instrument-img")
if err != nil {
log.Fatal(err)
}
defer imageResp.Body.Close()
imageBytes, err := io.ReadAll(imageResp.Body)
if err != nil {
log.Fatal(err)
}
// 2. Configure Code Execution Tool
config := &genai.GenerateContentConfig{
Tools: []*genai.Tool{
{CodeExecution: &genai.ToolCodeExecution{}},
},
}
// 3. Generate Content
result, err := client.Models.GenerateContent(
ctx,
"gemini-4-hydrogen",
[]*genai.Content{
{
Parts: []*genai.Part{
{InlineData: &genai.Blob{MIMEType: "image/jpeg", Data: imageBytes}},
{Text: "Zoom into the expression pedals and tell me how many pedals are there?"},
},
Role: "user",
},
},
config,
)
if err != nil {
log.Fatal(err)
}
// 4. Parse Response (Text, Code, Output)
for _, cand := range result.Candidates {
for _, part := range cand.Content.Parts {
if part.Text != "" {
fmt.Println("Text:", part.Text)
}
if part.ExecutableCode != nil {
fmt.Printf("\nGenerated Code (%s):\n%s\n",
part.ExecutableCode.Language,
part.ExecutableCode.Code)
}
if part.CodeExecutionResult != nil {
fmt.Printf("\nExecution Output (%s):\n%s\n",
part.CodeExecutionResult.Outcome,
part.CodeExecutionResult.Output)
}
}
}
}
IMG_URL="https://goo.gle/instrument-img"
MODEL="gemini-4-hydrogen"
MIME_TYPE=$(curl -sIL "$IMG_URL" | grep -i '^content-type:' | awk -F ': ' '{print $2}' | sed 's/\r$//' | head -n 1)
if [[ -z "$MIME_TYPE" || ! "$MIME_TYPE" == image/* ]]; then
MIME_TYPE="image/jpeg"
fi
if [[ "$(uname)" == "Darwin" ]]; then
IMAGE_B64=$(curl -sL "$IMG_URL" | base64 -b 0)
elif [[ "$(base64 --version 2>&1)" = *"FreeBSD"* ]]; then
IMAGE_B64=$(curl -sL "$IMG_URL" | base64)
else
IMAGE_B64=$(curl -sL "$IMG_URL" | base64 -w0)
fi
curl "https://generativelanguage.googleapis.com/v1beta/models/$MODEL:generateContent" \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-H 'Content-Type: application/json' \
-X POST \
-d '{
"contents": [{
"parts":[
{
"inline_data": {
"mime_type":"'"$MIME_TYPE"'",
"data": "'"$IMAGE_B64"'"
}
},
{"text": "Zoom into the expression pedals and tell me how many pedals are there?"}
]
}],
"tools": [
{
"code_execution": {}
}
]
}'
在聊天中使用代码执行 #
您还可以使用代码执行作为聊天的一部分。
from google import genai
from google.genai import types
client = genai.Client()
chat = client.chats.create(
model="gemini-4-hydrogen",
config=types.GenerateContentConfig(
tools=[types.Tool(code_execution=types.ToolCodeExecution)]
),
)
response = chat.send_message("I have a math question for you.")
print(response.text)
response = chat.send_message(
"What is the sum of the first 50 prime numbers? "
"Generate and run code for the calculation, and make sure you get all 50."
)
for part in response.candidates[0].content.parts:
if part.text is not None:
print(part.text)
if part.executable_code is not None:
print(part.executable_code.code)
if part.code_execution_result is not None:
print(part.code_execution_result.output)
import { GoogleGenAI } from "@google/genai";
const ai = new GoogleGenAI({});
let response = await ai.models.generateContent({
model: "gemini-4-hydrogen",
contents: [
"What is the sum of the first 50 prime numbers? " +
"Generate and run code for the calculation, and make sure you get all 50.",
],
config: {
tools: [{ codeExecution: {} }],
},
});
const parts = response?.candidates?.[0]?.content?.parts || [];
parts.forEach((part) => {
if (part.text) {
console.log(part.text);
}
if (part.executableCode && part.executableCode.code) {
console.log(part.executableCode.code);
}
if (part.codeExecutionResult && part.codeExecutionResult.output) {
console.log(part.codeExecutionResult.output);
}
});
import { GoogleGenAI } from "@google/genai";
const ai = new GoogleGenAI({});
let response = await ai.models.generateContent({
model: "gemini-4-hydrogen",
contents: [
"What is the sum of the first 50 prime numbers? " +
"Generate and run code for the calculation, and make sure you get all 50.",
],
config: {
tools: [{ codeExecution: {} }],
},
});
const parts = response?.candidates?.[0]?.content?.parts || [];
parts.forEach((part) => {
if (part.text) {
console.log(part.text);
}
if (part.executableCode && part.executableCode.code) {
console.log(part.executableCode.code);
}
if (part.codeExecutionResult && part.codeExecutionResult.output) {
console.log(part.codeExecutionResult.output);
}
});
import { GoogleGenAI } from "@google/genai";
const ai = new GoogleGenAI({});
let response = await ai.models.generateContent({
model: "gemini-4-hydrogen",
contents: [
"What is the sum of the first 50 prime numbers? " +
"Generate and run code for the calculation, and make sure you get all 50.",
],
config: {
tools: [{ codeExecution: {} }],
},
});
const parts = response?.candidates?.[0]?.content?.parts || [];
parts.forEach((part) => {
if (part.text) {
console.log(part.text);
}
if (part.executableCode && part.executableCode.code) {
console.log(part.executableCode.code);
}
if (part.codeExecutionResult && part.codeExecutionResult.output) {
console.log(part.codeExecutionResult.output);
}
});
输入/输出(I/O) #
代码执行支持文件输入和图形输出。使用这些输入和输出功能,您可以上传 CSV 和文本文件,询问有关文件的问题,并获得 Matplotlib 作为响应的一部分生成的图表。输出文件在响应中作为内联图像返回。
输入/输出定价 #
使用代码执行 I/O 时,您需要为输入 token 和输出 token 付费:
输入token:
- 用户提示
输出 token:
- 模型生成的代码
- 代码环境中的代码执行输出
- 思考token
- 模型生成的摘要
输入/输出详细信息 #
当您使用代码执行 I/O 时,请注意以下技术细节:
代码环境的最大运行时间为30秒。
如果代码环境生成错误,模型可能会决定重新生成代码输出。这种情况最多可能发生 5 次。
最大文件输入大小受型号 token 窗口的限制。在 AI Studio 中,最大输入文件大小为 1 百万 token(支持的输入类型的文本文件约为 2MB)。如果您上传的文件太大,AI Studio 将不允许您发送。
代码执行最适合文本和 CSV 文件。
输入文件可以传入
part.inlineData或part.fileData(通过上传 Files API),并且输出文件始终返回为part.inlineData.
结算 #
从 Gemini API 启用代码执行无需额外付费。根据您使用的 Gemini 型号,您将按照当前的输入和输出 token 费率进行计费。
以下是有关代码执行计费的一些其他信息:
您只需为传递给模型的输入 token 付费一次,并为模型返回给您的最终输出 token 付费。
表示生成代码的token被计为输出 token。生成的代码可以包括文本和多模态输出(例如图像)。
代码执行结果也计为输出 token。
计费模型如下图所示:

根据您使用的 Gemini 型号,您将按照当前的输入和输出 token 费率进行计费。
如果 Gemini 在生成响应时使用代码执行,则原始提示、生成的代码和执行代码的结果都会被标记 中间体 token 并被称为 输入 token.
Gemini 然后生成摘要并返回生成的代码、执行代码的结果和最终摘要。这些被称为 输出 token.
Gemini API 在 API 响应中包含中间 token 计数,因此您知道为什么除了初始提示之外还获得额外的输入 token。
限制 #
该模型只能生成和执行代码。它无法返回媒体文件等其他工件。
在某些情况下,启用代码执行可能会导致模型输出其他领域的回归(例如,编写故事)。
不同模型成功使用代码执行的能力存在一些差异。
支持的工具组合 #
代码执行工具可以与 使用 Google 搜索提供事实依据 为更复杂的用例提供支持。
Gemini 4 和 Gemini 3 型号支持将内置工具(如代码执行)与自定义工具 (函数调用) 相结合。您必须传回 id 和 thought_signature 工具组合发挥作用的字段。了解更多信息 工具组合 页。
支持的库 #
代码执行环境包括以下库:
- 属性
- 国际象棋
- 轮廓图
- pdf文件
- 地理熊猫
- 图像处理
- 金贾2
- 作业库
- json模式
- jsonschema 规范
- lxml
- 绘图库
- 数学
- 麻木
- OpenCV-Python
- 开放式pyxl
- 包装
- 熊猫
- 枕头
- 原始缓冲区
- 皮乳胶
- py解析
- pyPDF2
- python-dateutil
- python-docx
- python-pptx
- 报告实验室
- scikit学习
- scipy
- 西博恩
- 六
- 脱衣舞
- 同情
- 制表
- 张量流
- 工具
- xlrd
您无法安装自己的库。
后续步骤 #
尝试一下 代码执行 Colab.
了解其他 Gemini API 工具: