Gemini 3 全维度技术解析：从认知到落地实战指南

pip install google-generativeai

npm install @google/generative-ai

import google.generativeai as genai

# 初始化客户端

genai.configure(api_key="YOUR_API_KEY")

# 配置生成参数

generation_config = {

"thinking_level": "low", # 简单任务，低推理深度

"temperature": 0.3,

"max_output_tokens": 1000,

}

# 创建模型实例

model = genai.GenerativeModel(

model_name="gemini-3-pro-preview",

generation_config=generation_config,

)

# 发送请求

prompt = "解释什么是微服务架构，核心优势是什么？"

response = model.generate_content(prompt)

# 处理响应

if response.result:

print("生成结果：")

print(response.text)

else:

print("请求失败：", response.prompt_feedback)

import { GoogleGenerativeAI } from "@google/generative-ai";

const ai = new GoogleGenerativeAI("YOUR_API_KEY");

async function getStructuredData() {

const model = ai.getGenerativeModel({

model: "gemini-3-pro-preview",

generationConfig: {

thinking_level: "medium",

response_mime_type: "application/json", // 指定JSON输出

},

});

const prompt = `

分析以下产品需求，输出JSON格式的功能清单：

需求：开发一款健身追踪App，支持记录运动类型、时长、消耗卡路里，生成周报表，提供饮食建议。

JSON格式要求：

{

"核心功能": [],

"辅助功能": [],

"技术需求": []

}

`;

const result = await model.generateContent(prompt);

const jsonResult = JSON.parse(result.response.text());

console.log("结构化功能清单：", jsonResult);

return jsonResult;

}

// 执行函数

getStructuredData().catch(console.error);

curl "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-preview:generateContent" \

-H "X-Goog-Api-Key: YOUR_API_KEY" \

-H "Content-Type: application/json" \

-X POST \

-d '{

"contents": [

{

"parts": [

{

"text": "以下是一份10万字的技术文档，请完成：1. 提取核心知识点；2. 生成目录结构；3. 识别潜在风险点。文档内容：[此处替换为长文档文本]"

}

]

}

],

"generationConfig": {

"thinking_level": "high",

"max_output_tokens": 64000

}

}'

import google.generativeai as genai

from PIL import Image

import os

# 初始化客户端

genai.configure(api_key="YOUR_API_KEY")

# 加载图片（支持本地文件或URL）

def load_image(image_path):

if image_path.startswith("http"):

return genai.upload_file(image_path)

else:

return Image.open(image_path)

# 配置多模态参数

generation_config = {

"thinking_level": "high",

"media_resolution": "hi", # 高清图片分析

"max_output_tokens": 2000,

}

# 创建模型

model = genai.GenerativeModel(

model_name="gemini-3-pro-image-preview",

generation_config=generation_config,

)

# 处理图片+文本混合输入

image = load_image("product_design.png") # 产品设计图

prompt = "分析这张产品设计图的UI风格，识别交互组件，指出可优化点"

response = model.generate_content([prompt, image])

# 输出结果

print("图片分析报告：")

print(response.text)

# 保存结果到文件

with open("image_analysis_report.md", "w", encoding="utf-8") as f:

f.write(response.text)

import { GoogleGenerativeAI } from "@google/generative-ai";

import fs from "fs";

const ai = new GoogleGenerativeAI("YOUR_API_KEY");

async function processVideo(videoPath) {

// 上传视频文件（支持MP4格式，建议时长）

const videoFile = await genai.uploadFile(videoPath, {

mimeType: "video/mp4",

displayName: "meeting_video"

});

const model = ai.getGenerativeModel({

model: "gemini-3-pro-preview",

generationConfig: {

thinking_level: "high",

media_resolution: "medium",

},

});

const prompt = `

分析这个会议视频，完成以下任务：

1. 提取关键讨论点（按时间轴排列）

2. 识别行动项与负责人

3. 生成结构化会议纪要（Markdown格式）

`;

const result = await model.generateContent([prompt, videoFile]);

const transcript = result.response.text();

// 保存会议纪要

fs.writeFileSync("meeting_minutes.md", transcript, "utf-8");

console.log("会议纪要已生成");

return transcript;

}

// 执行视频处理

processVideo("team_meeting.mp4").catch(console.error);

import google.generativeai as genai

genai.configure(api_key="YOUR_API_KEY")

# 待检测的多线程C++代码

cpp_code = """

#include >

#include

#include int counter = 0;

void increment() {

for (int i = 0; i 10000; ++i) {

counter++; // 潜在的竞态条件

}

}

int main() {

std::vector;

for (int i = 0; i ) {

threads.emplace_back(increment);

}

for (auto& t : threads) {

t.join();

}

std::cout <: " << counter <::endl;

return 0;

}

"""

model = genai.GenerativeModel(

model_name="gemini-3-pro-preview",

generation_config={

"thinking_level": "high", # 编码任务需要深度推理

"temperature": 0.1,

},

)

prompt = f"找出以下C++代码中的竞态条件，并提供修复方案：\n{cpp_code}"

response = model.generate_content(prompt)

print("代码分析结果：")

print(response.text)

import { GoogleGenerativeAI } from "@google/generative-ai";

const ai = new GoogleGenerativeAI("YOUR_API_KEY");

async function generateTests() {

const model = ai.getGenerativeModel({

model: "gemini-3-pro-preview",

generation_config: {

thinking_level: "high",

response_mime_type: "application/javascript",

},

});

// 待测试的函数

const targetFunction = `

// 计算两个数的最大公约数

function gcd(a, b) {

while (b !== 0) {

let temp = b;

b = a % b;

a = temp;

}

return a;

}

`;

const prompt = `

为以下JavaScript函数生成完整的Jest测试用例，包括：

1. 正常输入场景

2. 边界值场景（0、负数、大数）

3. 异常输入处理

函数代码：${targetFunction}

`;

const result = await model.generateContent(prompt);

const testCode = result.response.text();

// 保存测试文件

fs.writeFileSync("gcd.test.js", testCode, "utf-8");

console.log("测试用例已生成");

return testCode;

}

generateTests().catch(console.error);

import google.generativeai as genai

import requests

genai.configure(api_key="YOUR_API_KEY")

# 定义外部工具（天气查询API）

def get_weather(city):

weather_api_key = "YOUR_WEATHER_API_KEY"

url = f"https://api.openweathermap.org/data/2.5/weather?q={city}&appid={weather_api_key}&units=metric"

response = requests.get(url)

return response.json()

# 配置工具调用

tools = [

{

"name": "get_weather",

"description": "查询指定城市的天气信息",

"parameters": {

"type": "object",

"properties": {

"city": {

"type": "string",

"description": "城市名称"

}

},

"required": ["city"]

}

}

]

# 创建模型

model = genai.GenerativeModel(

model_name="gemini-3-pro-preview",

tools=tools,

generation_config={

"thinking_level": "medium",

},

)

# 启动Agent对话

conversation = model.start_chat()

prompt = "帮我查询北京明天的天气，然后推荐合适的户外活动"

response = conversation.send_message(prompt)

# 处理工具调用请求

if response.candidates[0].content.parts[0].function_call:

func_call = response.candidates[0].content.parts[0].function_call

if func_call.name == "get_weather":

# 执行工具调用

city = func_call.args["city"]

weather_data = get_weather(city)

# 将工具返回结果反馈给模型

feedback_prompt = f"天气查询结果：{weather_data}"

final_response = conversation.send_message(feedback_prompt)

print("Agent最终推荐：")

print(final_response.text)

import google.generativeai as genai

import pandas as pd

from datetime import datetime

genai.configure(api_key="YOUR_API_KEY")

# 定义业务工具

def fetch_sales_data(start_date, end_date):

"""获取指定日期范围内的销售数据"""

# 实际场景中对接数据库或API

data = pd.DataFrame({

"日期": pd.date_range(start=start_date, end=end_date),

"销售额": [12000, 15000, 13500, 14200, 16800],

"订单数": [320, 380, 350, 365, 410]

})

return data.to_dict()

def generate_report(data, report_type="markdown"):

"""生成销售报表"""

df = pd.DataFrame(data)

total_sales = df["销售额"].sum()

avg_orders = df["订单数"].mean()

if report_type == "markdown":

return f"""

# 销售报表（{df["日期"].min()} 至 {df["日期"].max()}）

| 指标 | 数值 |

|------|------|

| 总销售额 | {total_sales:,} 元 |

| 平均订单数 | {avg_orders:.0f} |

| 最高单日销售额 | {df["销售额"].max():,} 元 |

"""

return {"总销售额": total_sales, "平均订单数": avg_orders}

# 配置Agent工具

tools = [

{

"name": "fetch_sales_data",

"description": "获取指定日期范围内的销售数据",

"parameters": {

"type": "object",

"properties": {

"start_date": {"type": "string", "format": "date"},

"end_date": {"type": "string", "format": "date"}

},

"required": ["start_date", "end_date"]

}

},

{

"name": "generate_report",

"description": "根据销售数据生成报表",

"parameters": {

"type": "object",

"properties": {

"data": {"type": "object", "description": "销售数据字典"},

"report_type": {"type": "string", "enum": ["markdown", "json"]}

},

"required": ["data"]

}

}

]

# 初始化Agent

model = genai.GenerativeModel(

model_name="gemini-3-pro-preview",

tools=tools,

generation_config={"thinking_level": "high"}

)

chat = model.start_chat()

# 发送任务指令

prompt = f"生成2025年11月1日至2025年11月5日的销售报表，用Markdown格式输出"

response = chat.send_message(prompt)

# 处理工具调用链

while True:

if not response.candidates[0].content.parts[0].function_call:

break

func_call = response.candidates[0].content.parts[0].function_call

func_name = func_call.name

args = func_call.args

# 执行对应函数

if func_name == "fetch_sales_data":

result = fetch_sales_data(args["start_date"], args["end_date"])

elif func_name == "generate_report":

result = generate_report(args["data"], args.get("report_type", "markdown"))

else:

result = "未知函数"

# 反馈执行结果

response = chat.send_message(f"工具执行结果：{result}")

# 输出最终报表

print("生成的销售报表：")

print(response.text)

import google.generativeai as genai

from google.api_core.exceptions import GoogleAPIError, ResourceExhausted

genai.configure(api_key="YOUR_API_KEY")

def safe_generate_content(prompt, model_name="gemini-3-pro-preview", retries=3):

"""带重试机制的安全生成函数"""

model = genai.GenerativeModel(model_name)

for attempt in range(retries):

try:

response = model.generate_content(prompt)

# 检查内容安全反馈

if response.prompt_feedback.block_reason:

return f"请求被拒绝：{response.prompt_feedback.block_reason}"

return response.text

except ResourceExhausted as e:

# 配额耗尽，延迟重试

sleep_time = 2 ** attempt # 指数退避

print(f"配额耗尽，{sleep_time}秒后重试（第{attempt+1}次）")

time.sleep(sleep_time)

except GoogleAPIError as e:

print(f"API错误：{e.message}")

if attempt == retries - 1:

return "请求失败，请稍后重试"

except Exception as e:

print(f"未知错误：{str(e)}")

return "请求异常"

# 使用示例

result = safe_generate_content("复杂的技术文档生成任务...", retries=5)

print(result)