> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mixroute.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Gemini 原生（文字）

> 使用 Google Gemini 原生格式呼叫 API

## 簡介

Gemini 原生 API 採用 Google Gemini 的請求與回應格式，適用於 Google 官方客戶端（如 `google-generativeai` SDK）或需要直接使用 Gemini 資料結構的情境。

若使用 OpenAI 相容客戶端（如 OpenAI SDK），請使用 `/v1/chat/completions` 介面。

### 與 OpenAI 格式的差異

| 特性   | Gemini 原生            | OpenAI 相容                    |
| ---- | -------------------- | ---------------------------- |
| 訊息結構 | `contents[].parts[]` | `messages[].content`         |
| 角色名稱 | `user` / `model`     | `user` / `assistant`         |
| 串流參數 | URL 參數 `?alt=sse`    | Body 參數 `stream: true`       |
| 系統提示 | `systemInstruction`  | `messages[0].role: "system"` |
| 多模態  | `parts[]` 陣列混合       | `content[]` 陣列混合             |

## API 端點

| 功能           | 方法   | 路徑                                                     |
| ------------ | ---- | ------------------------------------------------------ |
| 文字生成（非串流）    | POST | `/v1beta/models/{model}:generateContent`               |
| 文字生成（串流）     | POST | `/v1beta/models/{model}:streamGenerateContent?alt=sse` |
| 單條 Embedding | POST | `/v1beta/models/{model}:embedContent`                  |
| 批次 Embedding | POST | `/v1beta/models/{model}:batchEmbedContents`            |

## 認證

支援兩種認證方式：

| 方式               | Header           | 範例                     |
| ---------------- | ---------------- | ---------------------- |
| Bearer Token（推薦） | `Authorization`  | `Bearer sk-xxxxxxxxxx` |
| Google 風格        | `x-goog-api-key` | `sk-xxxxxxxxxx`        |

## 請求參數

| 參數                  | 類型     | 必需 | 描述             |
| ------------------- | ------ | -- | -------------- |
| `contents`          | array  | 是  | 對話內容陣列         |
| `generationConfig`  | object | 否  | 生成設定參數         |
| `safetySettings`    | array  | 否  | 安全過濾設定         |
| `systemInstruction` | object | 否  | 系統指令           |
| `tools`             | array  | 否  | 工具定義（函式呼叫、搜尋等） |
| `cachedContent`     | string | 否  | 快取內容名稱         |

### generationConfig 參數

| 參數                | 類型      | 描述           |
| ----------------- | ------- | ------------ |
| `temperature`     | number  | 隨機性（0-2）     |
| `topP`            | number  | 核取樣（0-1）     |
| `topK`            | integer | Top-K 取樣     |
| `maxOutputTokens` | integer | 最大輸出 token 數 |
| `stopSequences`   | array   | 停止序列         |
| `candidateCount`  | integer | 候選回應數量       |
| `thinkingConfig`  | object  | 思考模式設定       |

## 基礎範例

<Tabs>
  <Tab title="cURL（非串流）">
    ```bash theme={null}
    curl -X POST "https://api.mixroute.ai/v1beta/models/gemini-2.5-pro:generateContent" \
      -H "Content-Type: application/json" \
      -H "Authorization: Bearer sk-xxxxxxxxxx" \
      -d '{
        "contents": [
          {"role": "user", "parts": [{"text": "用一句話介紹人工智慧"}]}
        ],
        "generationConfig": {
          "temperature": 0.7,
          "maxOutputTokens": 1024
        }
      }'
    ```
  </Tab>

  <Tab title="cURL（串流）">
    ```bash theme={null}
    curl -X POST "https://api.mixroute.ai/v1beta/models/gemini-2.5-pro:streamGenerateContent?alt=sse" \
      -H "Content-Type: application/json" \
      -H "Authorization: Bearer sk-xxxxxxxxxx" \
      -d '{
        "contents": [
          {"role": "user", "parts": [{"text": "寫一首關於春天的詩"}]}
        ],
        "generationConfig": {
          "temperature": 0.8,
          "maxOutputTokens": 2048
        }
      }'
    ```
  </Tab>

  <Tab title="Python">
    ```python theme={null}
    import google.generativeai as genai

    genai.configure(
        api_key="sk-xxxxxxxxxx",
        transport="rest",
        client_options={"api_endpoint": "https://api.mixroute.ai"}
    )

    model = genai.GenerativeModel("gemini-2.5-pro")
    response = model.generate_content("用一句話介紹人工智慧")
    print(response.text)
    ```
  </Tab>

  <Tab title="Node.js">
    ```javascript theme={null}
    import { GoogleGenerativeAI } from "@google/generative-ai";

    const genAI = new GoogleGenerativeAI("sk-xxxxxxxxxx");

    // 自訂端點
    const model = genAI.getGenerativeModel(
      { model: "gemini-2.5-pro" },
      { baseUrl: "https://api.mixroute.ai/v1beta" }
    );

    const result = await model.generateContent("用一句話介紹人工智慧");
    console.log(result.response.text());
    ```
  </Tab>
</Tabs>

## 進階功能

<Tabs>
  <Tab title="思考模式">
    ### 思考模式（Thinking）

    Gemini 2.5 Pro 和 Gemini 3 Pro 支援思考模式，使模型在回答前進行深度推理。

    **Gemini 2.5 Pro** - 使用 `thinkingBudget`：

    ```json theme={null}
    {
      "contents": [{"role": "user", "parts": [{"text": "給一道幾何題並分步解析"}]}],
      "generationConfig": {
        "maxOutputTokens": 16384,
        "thinkingConfig": {
          "includeThoughts": true,
          "thinkingBudget": 8192
        }
      }
    }
    ```

    **Gemini 3 Pro** - 使用 `thinkingLevel`：

    ```json theme={null}
    {
      "contents": [{"role": "user", "parts": [{"text": "解釋量子糾纏的原理"}]}],
      "generationConfig": {
        "maxOutputTokens": 16384,
        "thinkingConfig": {
          "includeThoughts": true,
          "thinkingLevel": "MEDIUM"
        }
      }
    }
    ```

    | 參數               | 適用模型           | 可選值                       |
    | ---------------- | -------------- | ------------------------- |
    | `thinkingBudget` | Gemini 2.5 Pro | 1-24576（token 數量）         |
    | `thinkingLevel`  | Gemini 3 Pro   | `LOW` / `MEDIUM` / `HIGH` |
  </Tab>

  <Tab title="多模態輸入">
    ### 多模態輸入

    支援圖像、音訊、影片等多種輸入格式。

    **圖像輸入（Base64）：**

    ```json theme={null}
    {
      "contents": [
        {
          "role": "user",
          "parts": [
            {
              "inlineData": {
                "mimeType": "image/jpeg",
                "data": "BASE64_ENCODED_IMAGE"
              }
            },
            {"text": "描述這張圖片的內容"}
          ]
        }
      ]
    }
    ```

    **圖像輸入（URL）：**

    ```json theme={null}
    {
      "contents": [
        {
          "role": "user",
          "parts": [
            {
              "fileData": {
                "mimeType": "image/jpeg",
                "fileUri": "https://example.com/image.jpg"
              }
            },
            {"text": "這張圖片裡有什麼？"}
          ]
        }
      ]
    }
    ```

    **支援的 MIME 類型：**

    * 圖像：`image/jpeg`, `image/png`, `image/gif`, `image/webp`
    * 音訊：`audio/mp3`, `audio/wav`, `audio/aac`
    * 影片：`video/mp4`, `video/webm`
    * 文件：`application/pdf`
  </Tab>

  <Tab title="工具呼叫">
    ### 工具呼叫（Function Calling）

    ```json theme={null}
    {
      "contents": [{"role": "user", "parts": [{"text": "上海今天天氣怎麼樣？"}]}],
      "tools": [
        {
          "functionDeclarations": [
            {
              "name": "get_weather",
              "description": "取得指定城市的天氣資訊",
              "parameters": {
                "type": "object",
                "properties": {
                  "location": {
                    "type": "string",
                    "description": "城市名稱"
                  },
                  "unit": {
                    "type": "string",
                    "enum": ["celsius", "fahrenheit"],
                    "description": "溫度單位"
                  }
                },
                "required": ["location"]
              }
            }
          ]
        }
      ],
      "toolConfig": {
        "functionCallingConfig": {
          "mode": "AUTO"
        }
      }
    }
    ```

    **工具呼叫模式：**

    * `AUTO`：模型自動決定是否呼叫
    * `ANY`：強制呼叫工具
    * `NONE`：禁止呼叫工具
  </Tab>

  <Tab title="Google 搜尋">
    ### Google 搜尋（Grounding）

    啟用 Google 搜尋取得即時資訊：

    ```json theme={null}
    {
      "contents": [{"role": "user", "parts": [{"text": "今天北京天氣怎麼樣？"}]}],
      "tools": [
        {
          "googleSearch": {}
        }
      ]
    }
    ```

    **動態檢索設定：**

    ```json theme={null}
    {
      "contents": [{"role": "user", "parts": [{"text": "最新的 AI 新聞"}]}],
      "tools": [
        {
          "googleSearch": {
            "dynamicRetrievalConfig": {
              "mode": "MODE_DYNAMIC",
              "dynamicThreshold": 0.5
            }
          }
        }
      ]
    }
    ```
  </Tab>

  <Tab title="串流輸出">
    ### 串流輸出

    **Python 串流：**

    ```python theme={null}
    import google.generativeai as genai

    genai.configure(
        api_key="sk-xxxxxxxxxx",
        transport="rest",
        client_options={"api_endpoint": "https://api.mixroute.ai"}
    )

    model = genai.GenerativeModel("gemini-2.5-pro")
    response = model.generate_content(
        "寫一篇關於人工智慧的文章",
        stream=True
    )

    for chunk in response:
        print(chunk.text, end="", flush=True)
    ```

    **cURL 串流：**

    ```bash theme={null}
    curl -X POST "https://api.mixroute.ai/v1beta/models/gemini-2.5-pro:streamGenerateContent?alt=sse" \
      -H "Content-Type: application/json" \
      -H "Authorization: Bearer sk-xxxxxxxxxx" \
      -d '{
        "contents": [{"role": "user", "parts": [{"text": "講一個故事"}]}]
      }'
    ```
  </Tab>

  <Tab title="Context Caching">
    ### Context Caching（上下文快取）

    對於長文本或多輪對話，使用快取可以節省 token 消耗。

    **建立快取：**

    ```bash theme={null}
    curl -X POST "https://api.mixroute.ai/v1beta/cachedContents" \
      -H "Content-Type: application/json" \
      -H "Authorization: Bearer sk-xxxxxxxxxx" \
      -d '{
        "model": "models/gemini-2.5-pro",
        "displayName": "my-cache",
        "contents": [
          {
            "role": "user",
            "parts": [{"text": "這是一段很長的文件內容..."}]
          }
        ],
        "ttl": "3600s"
      }'
    ```

    **使用快取：**

    ```json theme={null}
    {
      "cachedContent": "cachedContents/abc123",
      "contents": [
        {"role": "user", "parts": [{"text": "基於上述文件，總結要點"}]}
      ]
    }
    ```
  </Tab>

  <Tab title="圖像生成">
    ### 圖像生成

    使用 Gemini 2.0 Flash 或 Imagen 模型生成圖像：

    ```json theme={null}
    {
      "contents": [
        {
          "role": "user",
          "parts": [{"text": "生成一張日落時分海邊的圖片"}]
        }
      ],
      "generationConfig": {
        "responseModalities": ["TEXT", "IMAGE"]
      }
    }
    ```

    **Imagen 模型：**

    ```bash theme={null}
    curl -X POST "https://api.mixroute.ai/v1beta/models/imagen-3.0-generate-002:predict" \
      -H "Content-Type: application/json" \
      -H "Authorization: Bearer sk-xxxxxxxxxx" \
      -d '{
        "instances": [
          {"prompt": "一隻可愛的貓咪在陽光下"}
        ],
        "parameters": {
          "sampleCount": 1
        }
      }'
    ```
  </Tab>
</Tabs>

## Embedding API

### 單條 Embedding

```bash theme={null}
curl -X POST "https://api.mixroute.ai/v1beta/models/text-embedding-004:embedContent" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-xxxxxxxxxx" \
  -d '{
    "content": {
      "parts": [{"text": "這是一段需要向量化的文字"}]
    }
  }'
```

**回應範例：**

```json theme={null}
{
  "embedding": {
    "values": [0.0123, -0.0456, 0.0789, ...]
  }
}
```

### 批次 Embedding

```bash theme={null}
curl -X POST "https://api.mixroute.ai/v1beta/models/text-embedding-004:batchEmbedContents" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-xxxxxxxxxx" \
  -d '{
    "requests": [
      {
        "model": "models/text-embedding-004",
        "content": {"parts": [{"text": "第一段文字"}]}
      },
      {
        "model": "models/text-embedding-004",
        "content": {"parts": [{"text": "第二段文字"}]}
      }
    ]
  }'
```

**回應範例：**

```json theme={null}
{
  "embeddings": [
    {"values": [0.0123, -0.0456, ...]},
    {"values": [0.0234, -0.0567, ...]}
  ]
}
```

## 回應格式

```json theme={null}
{
  "candidates": [
    {
      "content": {
        "parts": [{"text": "回覆文字"}],
        "role": "model"
      },
      "finishReason": "STOP",
      "safetyRatings": [
        {
          "category": "HARM_CATEGORY_HARASSMENT",
          "probability": "NEGLIGIBLE"
        }
      ]
    }
  ],
  "usageMetadata": {
    "promptTokenCount": 10,
    "candidatesTokenCount": 20,
    "totalTokenCount": 30
  }
}
```

## 錯誤處理

| HTTP 狀態碼 | 錯誤類型                | 描述          |
| -------- | ------------------- | ----------- |
| 400      | INVALID\_ARGUMENT   | 請求參數無效      |
| 401      | UNAUTHENTICATED     | API 金鑰無效或缺失 |
| 403      | PERMISSION\_DENIED  | 無權存取該模型     |
| 404      | NOT\_FOUND          | 模型不存在       |
| 429      | RESOURCE\_EXHAUSTED | 超出速率限制      |
| 500      | INTERNAL            | 伺服器內部錯誤     |

**錯誤回應範例：**

```json theme={null}
{
  "error": {
    "code": 400,
    "message": "Invalid value at 'contents[0].parts[0]'",
    "status": "INVALID_ARGUMENT"
  }
}
```

## 與 OpenAI 格式對比

| 特性       | Gemini 原生                        | OpenAI 相容                    |
| -------- | -------------------------------- | ---------------------------- |
| Base URL | `https://api.mixroute.ai/v1beta` | `https://api.mixroute.ai/v1` |
| 訊息結構     | `contents[].parts[]`             | `messages[].content`         |
| 角色名稱     | `user` / `model`                 | `user` / `assistant`         |
| 系統提示     | `systemInstruction`              | `messages[0].role: "system"` |
| 串流請求     | URL 參數 `?alt=sse`                | Body 參數 `stream: true`       |
| 溫度範圍     | 0-2                              | 0-2                          |
| 函式呼叫     | `tools[].functionDeclarations`   | `tools[].function`           |
| 搜尋增強     | `tools[].googleSearch`           | 不支援                          |
| 思考模式     | `thinkingConfig`                 | 不支援                          |

<RequestExample>
  ```bash cURL theme={null}
  curl --request POST \
    --url 'https://api.mixroute.ai/v1beta/models/gemini-2.5-pro:generateContent' \
    --header 'Authorization: Bearer sk-xxxxxxxxxx' \
    --header 'Content-Type: application/json' \
    --data '{
      "contents": [
        {"role": "user", "parts": [{"text": "用一句話介紹人工智慧"}]}
      ],
      "generationConfig": {
        "temperature": 0.7,
        "maxOutputTokens": 1024
      }
    }'
  ```

  ```python Python theme={null}
  import google.generativeai as genai

  genai.configure(
      api_key="sk-xxxxxxxxxx",
      transport="rest",
      client_options={"api_endpoint": "https://api.mixroute.ai"}
  )

  model = genai.GenerativeModel("gemini-2.5-pro")
  response = model.generate_content("用一句話介紹人工智慧")
  print(response.text)
  ```

  ```javascript JavaScript theme={null}
  import { GoogleGenerativeAI } from "@google/generative-ai";

  const genAI = new GoogleGenerativeAI("sk-xxxxxxxxxx");
  const model = genAI.getGenerativeModel(
    { model: "gemini-2.5-pro" },
    { baseUrl: "https://api.mixroute.ai/v1beta" }
  );

  const result = await model.generateContent("用一句話介紹人工智慧");
  console.log(result.response.text());
  ```

  ```php PHP theme={null}
  <?php
  $client = new GuzzleHttp\Client();
  $response = $client->post('https://api.mixroute.ai/v1beta/models/gemini-2.5-pro:generateContent', [
      'headers' => [
          'Authorization' => 'Bearer sk-xxxxxxxxxx',
          'Content-Type' => 'application/json',
      ],
      'json' => [
          'contents' => [
              ['role' => 'user', 'parts' => [['text' => '用一句話介紹人工智慧']]]
          ],
          'generationConfig' => [
              'temperature' => 0.7,
              'maxOutputTokens' => 1024
          ]
      ]
  ]);
  echo $response->getBody();
  ```

  ```go Go theme={null}
  package main

  import (
      "bytes"
      "encoding/json"
      "net/http"
  )

  func main() {
      payload := map[string]interface{}{
          "contents": []map[string]interface{}{
              {"role": "user", "parts": []map[string]string{{"text": "用一句話介紹人工智慧"}}},
          },
          "generationConfig": map[string]interface{}{
              "temperature": 0.7,
              "maxOutputTokens": 1024,
          },
      }
      body, _ := json.Marshal(payload)
      req, _ := http.NewRequest("POST", "https://api.mixroute.ai/v1beta/models/gemini-2.5-pro:generateContent", bytes.NewBuffer(body))
      req.Header.Set("Authorization", "Bearer sk-xxxxxxxxxx")
      req.Header.Set("Content-Type", "application/json")
      http.DefaultClient.Do(req)
  }
  ```

  ```java Java theme={null}
  import java.net.http.*;
  import java.net.URI;

  HttpClient client = HttpClient.newHttpClient();
  String json = """
      {
        "contents": [
          {"role": "user", "parts": [{"text": "用一句話介紹人工智慧"}]}
        ],
        "generationConfig": {
          "temperature": 0.7,
          "maxOutputTokens": 1024
        }
      }
      """;
  HttpRequest request = HttpRequest.newBuilder()
      .uri(URI.create("https://api.mixroute.ai/v1beta/models/gemini-2.5-pro:generateContent"))
      .header("Authorization", "Bearer sk-xxxxxxxxxx")
      .header("Content-Type", "application/json")
      .POST(HttpRequest.BodyPublishers.ofString(json))
      .build();
  HttpResponse<String> response = client.send(request, HttpResponse.BodyHandlers.ofString());
  System.out.println(response.body());
  ```

  ```ruby Ruby theme={null}
  require 'net/http'
  require 'json'

  uri = URI('https://api.mixroute.ai/v1beta/models/gemini-2.5-pro:generateContent')
  http = Net::HTTP.new(uri.host, uri.port)
  http.use_ssl = true

  request = Net::HTTP::Post.new(uri)
  request['Authorization'] = 'Bearer sk-xxxxxxxxxx'
  request['Content-Type'] = 'application/json'
  request.body = {
    contents: [
      { role: 'user', parts: [{ text: '用一句話介紹人工智慧' }] }
    ],
    generationConfig: {
      temperature: 0.7,
      maxOutputTokens: 1024
    }
  }.to_json

  response = http.request(request)
  puts response.body
  ```
</RequestExample>

<ResponseExample>
  ```json Response theme={null}
  {
    "candidates": [
      {
        "content": {
          "parts": [{"text": "人工智慧是一門研究如何讓電腦模擬和實現人類智慧的學科。"}],
          "role": "model"
        },
        "finishReason": "STOP",
        "index": 0,
        "safetyRatings": []
      }
    ],
    "usageMetadata": {
      "promptTokenCount": 10,
      "candidatesTokenCount": 20,
      "totalTokenCount": 30
    },
    "modelVersion": "gemini-2.5-pro"
  }
  ```
</ResponseExample>