---
title: PDF 支援
url: https://platform.claude.com/docs/zh-TW/build-with-claude/pdf-support
description: 使用 Claude 處理 PDF：從您的文件中擷取文字、分析圖表並理解視覺內容。
featureMetadata:
  status: ga
  zdr:
    eligibility: eligible
    note: Excludes [Covered Models](https://platform.claude.com/docs/en/manage-claude/api-and-data-retention#model-specific-data-retention-requirements).
  supportedPlatforms:
    Claude API: ga
    Claude Platform on AWS: ga
    Amazon Bedrock: ga
    Google Cloud: ga
    Microsoft Foundry: ga
---

您可以向 Claude 詢問您所提供的 PDF 中的任何文字、圖片、圖表和表格。一些範例使用情境：

* 分析財務報告並理解圖表／表格
* 從法律文件中擷取關鍵資訊
* 協助文件翻譯
* 將文件資訊轉換為結構化格式

## 開始之前

### 檢查 PDF 需求

Claude 可處理任何標準 PDF。請確保您的請求大小符合以下需求：

| 需求        | 限制                                                                                      |
| --------- | --------------------------------------------------------------------------------------- |
| 最大請求大小    | 32 MB（[依平台而異](https://platform.claude.com/docs/zh-TW/api/overview#request-size-limits)） |
| 每個請求的最大頁數 | 600（當請求的上下文視窗小於 1M tokens 時為 100）                                                       |
| 格式        | 標準 PDF（無密碼／加密）                                                                          |

這兩項限制皆適用於整個請求負載，包括與 PDF 一同傳送的任何其他內容。對於大型 PDF，請考慮使用 [Files API](https://platform.claude.com/docs/zh-TW/build-with-claude/files) 上傳並透過 `file_id` 引用，以保持請求負載較小。

<Tip>
  內容密集的 PDF（許多小字體頁面、複雜表格或大量圖形）可能在達到頁數限制之前就填滿「context window」（上下文視窗）。包含大型 PDF 的請求也可能在達到頁數限制之前失敗，即使使用 Files API 也是如此。請嘗試將文件分割成多個部分；對於大型檔案，由於每一頁都會被當作影像處理，降低內嵌影像的取樣率也會有所幫助。
</Tip>

由於 PDF 支援仰賴 Claude 的視覺能力，因此它受到與其他視覺任務相同的[限制與注意事項](https://platform.claude.com/docs/zh-TW/build-with-claude/vision#limitations)約束。

### 支援的平台與模型

所有[現行模型](https://platform.claude.com/docs/zh-TW/models/overview)皆支援 PDF 處理。若要透過 Amazon Bedrock 的 Converse API 使用 PDF 支援，請參閱 [Amazon Bedrock PDF 支援](https://platform.claude.com/docs/zh-TW/build-with-claude/pdf-support#amazon-bedrock-pdf-support)。

### Amazon Bedrock PDF 支援

當透過 Converse API（屬於 [Amazon Bedrock 上的 Claude（Opus 4.6 及更早版本）](https://platform.claude.com/docs/zh-TW/build-with-claude/claude-on-amazon-bedrock-legacy)的一部分）使用 PDF 支援時，有兩種不同的文件處理模式：

<Note>
  **重要：** 若要在 Converse API 中使用 Claude 完整的視覺化 PDF 理解能力，您必須啟用「citations」（引用）。若未啟用引用，API 將退回為僅進行基本文字擷取。深入了解[使用引用](https://platform.claude.com/docs/zh-TW/build-with-claude/citations)。
</Note>

#### 文件處理模式

1. **Converse Document Chat**（原始模式 - 僅文字擷取）

   * 提供 PDF 的基本文字擷取
   * 無法分析 PDF 中的影像、圖表或視覺版面配置
   * 3 頁的 PDF 約使用 1,000 個 tokens
   * 未啟用引用時自動使用

2. **Claude PDF Chat**（新模式 - 完整視覺理解）

   * 提供 PDF 的完整視覺分析
   * 能夠理解並分析圖表、圖形、影像和視覺版面配置
   * 將每一頁同時以文字和影像處理，以獲得全面的理解
   * 3 頁的 PDF 約使用 7,000 個 tokens
   * **需要在 Converse API 中啟用引用**

#### 主要限制

* **Converse API：** 視覺化 PDF 分析需要啟用引用。目前沒有不使用引用即可進行視覺分析的選項（與 InvokeModel API 不同）。
* **InvokeModel API：** 提供對 PDF 處理的完整控制，不強制使用引用。

#### 常見問題

如果在使用 Converse API 時 Claude 看不到您 PDF 中的影像或圖表，您很可能需要啟用引用旗標。若未啟用，Converse 將退回為僅進行基本文字擷取。

<Note>
  這是 Converse API 的已知限制。對於需要在不使用引用的情況下進行視覺化 PDF 分析的應用程式，請考慮改用 InvokeModel API。
</Note>

<Note>
  純文字檔案（例如 .txt、.csv 或 .md）可直接用於 document 區塊：以 MIME 類型 `text/plain` 將其上傳至 Files API，並透過 `file_id` 引用。二進位格式（例如 .xlsx 或 .docx）不支援用於 document 區塊，必須先轉換為文字或 PDF。請參閱[處理其他檔案格式](https://platform.claude.com/docs/zh-TW/build-with-claude/files#working-with-other-file-formats)。
</Note>

## 使用 Claude 處理 PDF

### 傳送您的第一個 PDF 請求

從使用 Messages API 的簡單範例開始。您可以透過三種方式向 Claude 提供 PDF：

1. 以 URL 引用線上託管的 PDF
2. 在 `document` 內容區塊中以 base64 編碼的 PDF
3. 透過 [Files API](https://platform.claude.com/docs/zh-TW/build-with-claude/files) 的 `file_id`

<Note>
  在 Amazon Bedrock 和 Google Cloud 上，目前僅提供 base64 編碼的來源。在 Microsoft Foundry 上，託管於 Azure 的部署不支援 Files API。
</Note>

#### 選項 1：基於 URL 的 PDF 文件

最簡單的方法是直接從 URL 引用 PDF：

<CodeGroup>
  ```bash cURL
  curl https://api.anthropic.com/v1/messages \
    -H "content-type: application/json" \
    -H "x-api-key: $ANTHROPIC_API_KEY" \
    -H "anthropic-version: 2023-06-01" \
    -d '{
      "model": "claude-opus-5-5",
      "max_tokens": 1024,
      "messages": [{
          "role": "user",
          "content": [{
              "type": "document",
              "source": {
                  "type": "url",
                  "url": "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
              }
          },
          {
              "type": "text",
              "text": "What are the key findings in this document?"
          }]
      }]
  }'
  ```

  ```bash CLI
  ant messages create --transform content --format yaml <<'YAML'
  model: claude-opus-5-5
  max_tokens: 1024
  messages:
    - role: user
      content:
        - type: document
          source:
            type: url
            url: https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf
        - type: text
          text: What are the key findings in this document?
  YAML
  ```

  ```python Python
  client = anthropic.Anthropic()
  message = client.messages.create(
      model="claude-opus-5-5",
      max_tokens=1024,
      messages=[
          {
              "role": "user",
              "content": [
                  {
                      "type": "document",
                      "source": {
                          "type": "url",
                          "url": "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf",
                      },
                  },
                  {"type": "text", "text": "What are the key findings in this document?"},
              ],
          }
      ],
  )

  print(message.content)
  ```

  ```typescript TypeScript
  const anthropic = new Anthropic();

  const response = await anthropic.messages.create({
    model: "claude-opus-5-5",
    max_tokens: 1024,
    messages: [
      {
        role: "user",
        content: [
          {
            type: "document",
            source: {
              type: "url",
              url: "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
            }
          },
          {
            type: "text",
            text: "What are the key findings in this document?"
          }
        ]
      }
    ]
  });

  console.log(response);
  ```

  ```csharp C#
  var client = new AnthropicClient();

  // 建立含 URL 的文件區塊
  var documentParam = new DocumentBlockParam
  {
      Source = new UrlPdfSource
      {
          Url = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf",
      },
  };

  // 建立包含文件與文字內容區塊的訊息
  var message = await client.Messages.Create(new MessageCreateParams
  {
      Model = Model.ClaudeOpus5_5,
      MaxTokens = 1024,
      Messages =
      [
          new()
          {
              Role = Role.User,
              Content = new List<ContentBlockParam>
              {
                  documentParam,
                  new TextBlockParam("What are the key findings in this document?"),
              },
          },
      ],
  });

  Console.WriteLine(string.Join("\n", message.Content));
  ```

  ```go Go
  client := anthropic.NewClient()

  message, err := client.Messages.New(context.TODO(), anthropic.MessageNewParams{
  	Model:     anthropic.ModelClaudeOpus5_5,
  	MaxTokens: 1024,
  	Messages: []anthropic.MessageParam{
  		anthropic.NewUserMessage(
  			anthropic.NewDocumentBlock(anthropic.URLPDFSourceParam{
  				URL: "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf",
  			}),
  			anthropic.NewTextBlock("What are the key findings in this document?"),
  		),
  	},
  })
  if err != nil {
  	panic(err)
  }

  fmt.Printf("%+v\n", message.Content)
  ```

  ```java Java
  AnthropicClient client = AnthropicOkHttpClient.fromEnv();

  // 使用 URL 建立文件區塊
  DocumentBlockParam documentParam = DocumentBlockParam.builder()
    .source(
      UrlPdfSource.builder()
        .url(
          "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
        )
        .build()
    )
    .build();

  // 建立包含文件與文字內容區塊的訊息
  MessageCreateParams params = MessageCreateParams.builder()
    .model(Model.CLAUDE_OPUS_5_5)
    .maxTokens(1024)
    .addUserMessageOfBlockParams(
      List.of(
        ContentBlockParam.ofDocument(documentParam),
        ContentBlockParam.ofText(
          TextBlockParam.builder()
            .text("What are the key findings in this document?")
            .build()
        )
      )
    )
    .build();

  Message message = client.messages().create(params);
  System.out.println(message.content());
  ```

  ```php PHP
  $client = new Client();

  $message = $client->messages->create(
      maxTokens: 1024,
      messages: [
          [
              'role' => 'user',
              'content' => [
                  [
                      'type' => 'document',
                      'source' => [
                          'type' => 'url',
                          'url' => 'https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf',
                      ],
                  ],
                  [
                      'type' => 'text',
                      'text' => 'What are the key findings in this document?',
                  ],
              ],
          ],
      ],
      model: 'claude-opus-5-5',
  );

  echo $message;
  ```

  ```ruby Ruby
  anthropic = Anthropic::Client.new

  message = anthropic.messages.create(
    model: "claude-opus-5-5",
    max_tokens: 1024,
    messages: [
      {
        role: "user",
        content: [
          {
            type: "document",
            source: {
              type: "url",
              url: "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
            }
          },
          {type: "text", text: "What are the key findings in this document?"}
        ]
      }
    ]
  )

  puts(message.content)
  ```
</CodeGroup>

回應會在 `content` 中以文字區塊的形式傳回 Claude 的分析，並在 `usage` 中提供 token 消耗量：

```json Output
{
  "id": "msg_01Hfp8YuFjQ55VgWbpdHDehB",
  "type": "message",
  "role": "assistant",
  "model": "claude-opus-5-5",
  "content": [
    {
      "type": "text",
      "text": "This document is an addendum to the Claude 3 model card, reporting updated evaluation results. The key findings include..."
    }
  ],
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 45000,
    "output_tokens": 300
  }
}
```

#### 選項 2：Base64 編碼的 PDF 文件

如果您需要從本機系統傳送 PDF，或在沒有可用 URL 的情況下：

<CodeGroup>
  ```bash cURL
  # 方法 1：擷取並編碼遠端 PDF
  curl -sL "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf" | base64 | tr -d '\n' > pdf_base64.txt

  # 方法 2：編碼本機 PDF 檔案
  # base64 document.pdf | tr -d '\n' > pdf_base64.txt

  # 使用 pdf_base64.txt 的內容建立 JSON 請求檔案
  jq -n --rawfile PDF_BASE64 pdf_base64.txt '{
      "model": "claude-opus-5-5",
      "max_tokens": 1024,
      "messages": [{
          "role": "user",
          "content": [{
              "type": "document",
              "source": {
                  "type": "base64",
                  "media_type": "application/pdf",
                  "data": $PDF_BASE64
              }
          },
          {
              "type": "text",
              "text": "What are the key findings in this document?"
          }]
      }]
  }' > request.json

  # 使用該 JSON 檔案傳送 API 請求
  curl https://api.anthropic.com/v1/messages \
    -H "content-type: application/json" \
    -H "x-api-key: $ANTHROPIC_API_KEY" \
    -H "anthropic-version: 2023-06-01" \
    -d @request.json
  ```

  ```bash CLI
  ant messages create \
    --model claude-opus-5-5 \
    --max-tokens 1024 \
    --transform content \
    --format yaml <<'YAML'
  messages:
    - role: user
      content:
        - type: document
          source:
            type: base64
            media_type: application/pdf
            data: "@./document.pdf"
        - type: text
          text: What are the key findings in this document?
  YAML
  ```

  ```python Python
  import base64
  import httpx2

  # 首先，載入 PDF 並進行編碼
  pdf_url = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
  pdf_data = base64.standard_b64encode(
      httpx2.get(pdf_url, follow_redirects=True).content
  ).decode("utf-8")

  # 替代方式：從本機檔案載入
  # with open("document.pdf", "rb") as f:
  #     pdf_data = base64.standard_b64encode(f.read()).decode("utf-8")

  # 使用 base64 編碼傳送給 Claude
  client = anthropic.Anthropic()
  message = client.messages.create(
      model="claude-opus-5-5",
      max_tokens=1024,
      messages=[
          {
              "role": "user",
              "content": [
                  {
                      "type": "document",
                      "source": {
                          "type": "base64",
                          "media_type": "application/pdf",
                          "data": pdf_data,
                      },
                  },
                  {"type": "text", "text": "What are the key findings in this document?"},
              ],
          }
      ],
  )

  print(message.content)
  ```

  ```typescript TypeScript
  // 方法 1：擷取遠端 PDF 並進行編碼
  const pdfURL =
    "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf";
  const pdfResponse = await fetch(pdfURL);
  const arrayBuffer = await pdfResponse.arrayBuffer();
  const pdfBase64 = Buffer.from(arrayBuffer).toString("base64");

  // 方法 2：從本機檔案載入
  // import { readFile } from "node:fs/promises";
  // const pdfBase64 = (await readFile('document.pdf')).toString('base64');

  // 傳送含 base64 編碼 PDF 的 API 請求
  const anthropic = new Anthropic();
  const response = await anthropic.messages.create({
    model: "claude-opus-5-5",
    max_tokens: 1024,
    messages: [
      {
        role: "user",
        content: [
          {
            type: "document",
            source: {
              type: "base64",
              media_type: "application/pdf",
              data: pdfBase64
            }
          },
          {
            type: "text",
            text: "What are the key findings in this document?"
          }
        ]
      }
    ]
  });

  console.log(response);
  ```

  ```csharp C#
  var client = new AnthropicClient();

  // 方法 1：下載並編碼遠端 PDF
  var pdfUrl = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf";
  using var httpClient = new HttpClient();
  var pdfBase64 = Convert.ToBase64String(await httpClient.GetByteArrayAsync(pdfUrl));

  // 方法 2：從本機檔案載入
  // var pdfBase64 = Convert.ToBase64String(await File.ReadAllBytesAsync("document.pdf"));

  // 使用 base64 資料建立文件區塊
  var documentParam = new DocumentBlockParam
  {
      Source = new Base64PdfSource { Data = pdfBase64 },
  };

  // 建立包含文件與文字內容區塊的訊息
  var message = await client.Messages.Create(new MessageCreateParams
  {
      Model = Model.ClaudeOpus5_5,
      MaxTokens = 1024,
      Messages =
      [
          new()
          {
              Role = Role.User,
              Content = new List<ContentBlockParam>
              {
                  documentParam,
                  new TextBlockParam("What are the key findings in this document?"),
              },
          },
      ],
  });

  Console.WriteLine(string.Join("\n", message.Content));
  ```

  ```go Go
  // 首先，載入 PDF 並進行編碼
  pdfURL := "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
  resp, err := http.Get(pdfURL)
  if err != nil {
  	panic(err)
  }
  defer resp.Body.Close()
  pdfBytes, err := io.ReadAll(resp.Body)
  if err != nil {
  	panic(err)
  }
  pdfBase64 := base64.StdEncoding.EncodeToString(pdfBytes)

  // 替代方式：從本機檔案載入（請在 imports 中加入 "os"）
  // pdfBytes, err := os.ReadFile("document.pdf")
  // pdfBase64 := base64.StdEncoding.EncodeToString(pdfBytes)

  // 使用 base64 編碼傳送給 Claude
  client := anthropic.NewClient()
  message, err := client.Messages.New(context.TODO(), anthropic.MessageNewParams{
  	Model:     anthropic.ModelClaudeOpus5_5,
  	MaxTokens: 1024,
  	Messages: []anthropic.MessageParam{
  		anthropic.NewUserMessage(
  			anthropic.NewDocumentBlock(anthropic.Base64PDFSourceParam{
  				Data: pdfBase64,
  			}),
  			anthropic.NewTextBlock("What are the key findings in this document?"),
  		),
  	},
  })
  if err != nil {
  	panic(err)
  }

  fmt.Printf("%+v\n", message.Content)
  ```

  ```java Java
  AnthropicClient client = AnthropicOkHttpClient.fromEnv();

  // 方法 1：下載遠端 PDF 並進行編碼
  String pdfUrl =
    "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf";
  HttpClient httpClient = HttpClient.newBuilder().followRedirects(HttpClient.Redirect.NORMAL).build();
  HttpRequest request = HttpRequest.newBuilder().uri(URI.create(pdfUrl)).GET().build();

  HttpResponse<byte[]> response = httpClient.send(
    request,
    HttpResponse.BodyHandlers.ofByteArray()
  );
  String pdfBase64 = Base64.getEncoder().encodeToString(response.body());

  // 方法 2：從本機檔案載入
  // byte[] fileBytes = Files.readAllBytes(Path.of("document.pdf"));
  // String pdfBase64 = Base64.getEncoder().encodeToString(fileBytes);

  // 使用 base64 資料建立文件區塊
  DocumentBlockParam documentParam = DocumentBlockParam.builder()
    .source(Base64PdfSource.builder().data(pdfBase64).build())
    .build();

  // 建立包含文件與文字內容區塊的訊息
  MessageCreateParams params = MessageCreateParams.builder()
    .model(Model.CLAUDE_OPUS_5_5)
    .maxTokens(1024)
    .addUserMessageOfBlockParams(
      List.of(
        ContentBlockParam.ofDocument(documentParam),
        ContentBlockParam.ofText(
          TextBlockParam.builder()
            .text("What are the key findings in this document?")
            .build()
        )
      )
    )
    .build();

  Message message = client.messages().create(params);
  System.out.println(message.content());
  ```

  ```php PHP
  $client = new Client();

  // 首先，載入 PDF 並進行編碼
  $pdf_url = 'https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf';
  $pdf_data = base64_encode(file_get_contents($pdf_url));

  // 替代方式：從本機檔案載入
  // $pdf_data = base64_encode(file_get_contents('document.pdf'));

  // 使用 base64 編碼傳送給 Claude
  $message = $client->messages->create(
      maxTokens: 1024,
      messages: [
          [
              'role' => 'user',
              'content' => [
                  [
                      'type' => 'document',
                      'source' => [
                          'type' => 'base64',
                          'media_type' => 'application/pdf',
                          'data' => $pdf_data,
                      ],
                  ],
                  [
                      'type' => 'text',
                      'text' => 'What are the key findings in this document?',
                  ],
              ],
          ],
      ],
      model: 'claude-opus-5-5',
  );

  echo $message;
  ```

  ```ruby Ruby
  require "open-uri"

  # 首先，載入 PDF 並進行編碼
  pdf_url = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
  pdf_bytes = URI.open(pdf_url, "rb") { |f| f.read }
  pdf_data = [pdf_bytes].pack("m0") # Base64-encode without newlines

  # 替代方案：從本機檔案載入
  # pdf_data = [File.binread("document.pdf")].pack("m0")

  # 使用 base64 編碼傳送給 Claude
  anthropic = Anthropic::Client.new
  message = anthropic.messages.create(
    model: "claude-opus-5-5",
    max_tokens: 1024,
    messages: [
      {
        role: "user",
        content: [
          {
            type: "document",
            source: {
              type: "base64",
              media_type: "application/pdf",
              data: pdf_data
            }
          },
          {type: "text", text: "What are the key findings in this document?"}
        ]
      }
    ]
  )

  puts(message.content)
  ```
</CodeGroup>

#### 選項 3：Files API

對於您會重複使用的 PDF，或當您想避免編碼開銷時，請使用 [Files API](https://platform.claude.com/docs/zh-TW/build-with-claude/files)：

<CodeGroup>
  ```bash cURL
  # 首先，將您的 PDF 上傳至 Files API
  FILE_ID=$(curl -sS -X POST https://api.anthropic.com/v1/files \
    -H "x-api-key: $ANTHROPIC_API_KEY" \
    -H "anthropic-version: 2023-06-01" \
    -F "file=@document.pdf" | jq -r '.id')

  # 接著在您的訊息中使用傳回的 file_id
  curl https://api.anthropic.com/v1/messages \
    -H "content-type: application/json" \
    -H "x-api-key: $ANTHROPIC_API_KEY" \
    -H "anthropic-version: 2023-06-01" \
    -d '{
      "model": "claude-opus-5-5",
      "max_tokens": 1024,
      "messages": [{
        "role": "user",
        "content": [{
          "type": "document",
          "source": {
            "type": "file",
            "file_id": "'"$FILE_ID"'"
          }
        },
        {
          "type": "text",
          "text": "What are the key findings in this document?"
        }]
      }]
    }'
  ```

  ```bash CLI
  # 首先，將您的 PDF 上傳至 Files API
  FILE_ID=$(ant files upload \
    --file ./document.pdf \
    --transform id \
    --raw-output)

  # 接著在您的訊息中使用傳回的 file_id
  ant messages create \
    --transform content \
    --format yaml <<YAML
  model: claude-opus-5-5
  max_tokens: 1024
  messages:
    - role: user
      content:
        - type: document
          source:
            type: file
            file_id: $FILE_ID
        - type: text
          text: What are the key findings in this document?
  YAML
  ```

  ```python Python
  client = anthropic.Anthropic()

  # 上傳 PDF 檔案
  with open("/path/to/document.pdf", "rb") as f:
      file_upload = client.files.upload(file=("document.pdf", f, "application/pdf"))

  # 在訊息中使用已上傳的檔案
  message = client.messages.create(
      model="claude-opus-5-5",
      max_tokens=1024,
      messages=[
          {
              "role": "user",
              "content": [
                  {
                      "type": "document",
                      "source": {"type": "file", "file_id": file_upload.id},
                  },
                  {"type": "text", "text": "What are the key findings in this document?"},
              ],
          }
      ],
  )

  print(message.content)
  ```

  ```typescript TypeScript
  import Anthropic, { toFile } from "@anthropic-ai/sdk";
  import fs from "node:fs";

  const anthropic = new Anthropic();

  // 上傳 PDF 檔案
  const fileUpload = await anthropic.files.upload({
    file: await toFile(fs.createReadStream("/path/to/document.pdf"), undefined, {
      type: "application/pdf"
    })
  });

  // 在訊息中使用已上傳的檔案
  const response = await anthropic.messages.create({
    model: "claude-opus-5-5",
    max_tokens: 1024,
    messages: [
      {
        role: "user",
        content: [
          {
            type: "document",
            source: {
              type: "file",
              file_id: fileUpload.id
            }
          },
          {
            type: "text",
            text: "What are the key findings in this document?"
          }
        ]
      }
    ]
  });

  console.log(response);
  ```

  ```csharp C#
  var client = new AnthropicClient();

  // 上傳 PDF 檔案
  var fileUpload = await client.Files.Upload(new FileUploadParams
  {
      File = new BinaryContent
      {
          Stream = File.OpenRead("/path/to/document.pdf"),
          FileName = "document.pdf",
          ContentType = new("application/pdf"),
      },
  });

  // 在訊息中使用已上傳的檔案
  var message = await client.Messages.Create(new MessageCreateParams
  {
      Model = Model.ClaudeOpus5_5,
      MaxTokens = 1024,
      Messages =
      [
          new()
          {
              Role = Role.User,
              Content = new List<ContentBlockParam>
              {
                  new DocumentBlockParam
                  {
                      Source = new FileDocumentSource { FileID = fileUpload.ID },
                  },
                  new TextBlockParam("What are the key findings in this document?"),
              },
          },
      ],
  });

  Console.WriteLine(string.Join("\n", message.Content));
  ```

  ```go Go
  client := anthropic.NewClient()

  // 上傳 PDF 檔案
  pdfFile, err := os.Open("/path/to/document.pdf")
  if err != nil {
  	panic(err)
  }
  defer pdfFile.Close()

  fileUpload, err := client.Files.Upload(context.TODO(), anthropic.FileUploadParams{
  	File: anthropic.File(pdfFile, "document.pdf", "application/pdf"),
  })
  if err != nil {
  	panic(err)
  }

  // 在訊息中使用已上傳的檔案
  message, err := client.Messages.New(context.TODO(), anthropic.MessageNewParams{
  	Model:     anthropic.ModelClaudeOpus5_5,
  	MaxTokens: 1024,
  	Messages: []anthropic.MessageParam{
  		anthropic.NewUserMessage(
  			anthropic.NewDocumentBlock(anthropic.FileDocumentSourceParam{
  				FileID: fileUpload.ID,
  			}),
  			anthropic.NewTextBlock("What are the key findings in this document?"),
  		),
  	},
  })
  if err != nil {
  	panic(err)
  }

  fmt.Printf("%+v\n", message.Content)
  ```

  ```java Java
  AnthropicClient client = AnthropicOkHttpClient.fromEnv();

  // 上傳 PDF 檔案
  FileMetadata file = client
    .files()
    .upload(FileUploadParams.builder().file(Path.of("/path/to/document.pdf")).build());

  // 在訊息中使用已上傳的檔案
  MessageCreateParams params = MessageCreateParams.builder()
    .model(Model.CLAUDE_OPUS_5_5)
    .maxTokens(1024)
    .addUserMessageOfBlockParams(
      List.of(
        ContentBlockParam.ofDocument(
          DocumentBlockParam.builder().fileSource(file.id()).build()
        ),
        ContentBlockParam.ofText(
          TextBlockParam.builder()
            .text("What are the key findings in this document?")
            .build()
        )
      )
    )
    .build();

  Message message = client.messages().create(params);
  System.out.println(message.content());
  ```

  ```php PHP
  use Anthropic\Core\FileParam;

  $client = new Client();

  // 上傳 PDF 檔案
  $file_upload = $client->files->upload(
      file: FileParam::fromResource(fopen('/path/to/document.pdf', 'r'), contentType: 'application/pdf'),
  );

  // 在訊息中使用已上傳的檔案
  $message = $client->messages->create(
      maxTokens: 1024,
      messages: [
          [
              'role' => 'user',
              'content' => [
                  [
                      'type' => 'document',
                      'source' => [
                          'type' => 'file',
                          'fileID' => $file_upload->id,
                      ],
                  ],
                  [
                      'type' => 'text',
                      'text' => 'What are the key findings in this document?',
                  ],
              ],
          ],
      ],
      model: 'claude-opus-5-5',
  );

  echo $message;
  ```

  ```ruby Ruby
  anthropic = Anthropic::Client.new

  # 上傳 PDF 檔案
  file_upload = File.open("/path/to/document.pdf", "rb") do |f|
    anthropic.files.upload(
      file: Anthropic::FilePart.new(f, filename: "document.pdf", content_type: "application/pdf")
    )
  end

  # 在訊息中使用已上傳的檔案
  message = anthropic.messages.create(
    model: "claude-opus-5-5",
    max_tokens: 1024,
    messages: [
      {
        role: "user",
        content: [
          {
            type: "document",
            source: {type: "file", file_id: file_upload.id}
          },
          {type: "text", text: "What are the key findings in this document?"}
        ]
      }
    ]
  )

  puts(message.content)
  ```
</CodeGroup>

### PDF 支援的運作方式

當您將 PDF 傳送給 Claude 時，會發生以下步驟：

<Steps>
  <Step title="系統擷取文件的內容。">
    * 系統將文件的每一頁轉換為影像。
    * 每一頁的文字會被擷取出來，並與該頁的影像一同提供。
  </Step>

  <Step title="Claude 同時分析文字與影像，以更好地理解文件。">
    * 文件以文字與影像組合的形式提供以供分析。
    * 這讓使用者能夠針對 PDF 的視覺元素（例如圖表、示意圖及其他非文字內容）尋求見解。
  </Step>

  <Step title="Claude 做出回應，並在相關時引用 PDF 的內容。">
    Claude 在回應時可以同時引用文字與視覺內容。您可以透過將 PDF 支援與以下功能整合來進一步提升效能：

    * [使用提示快取](https://platform.claude.com/docs/zh-TW/build-with-claude/pdf-support#use-prompt-caching)：提升重複分析的效能。
    * [處理文件批次](https://platform.claude.com/docs/zh-TW/build-with-claude/pdf-support#process-document-batches)：用於大量文件處理。
    * [工具使用](https://platform.claude.com/docs/zh-TW/agents-and-tools/tool-use/overview)：從文件中擷取特定資訊以作為工具輸入。
  </Step>
</Steps>

### 估算您的成本

PDF 檔案的 token 數量取決於從文件中擷取的文字總量以及頁數：

* 文字 token 成本：每頁通常使用 1,500–3,000 個 tokens，視內容密度而定。適用標準 API 定價，無額外 PDF 費用。
* 影像 token 成本：由於每一頁都會轉換為影像，因此適用相同的[基於影像的成本計算](https://platform.claude.com/docs/zh-TW/build-with-claude/vision#evaluate-image-size)。

您可以使用 [token 計數](https://platform.claude.com/docs/zh-TW/build-with-claude/token-counting)來估算您特定 PDF 的成本。

## 最佳化 PDF 處理

### 提升效能

請遵循以下最佳實務以獲得最佳結果：

* 在請求中將 PDF 放在文字之前
* 使用標準字型
* 確保文字清晰易讀
* 將頁面旋轉至正確的直立方向
* 在提示中使用邏輯頁碼（來自 PDF 檢視器）
* 必要時將大型 PDF 分割成多個區塊
* 針對重複分析啟用提示快取

### 擴展您的實作

對於大量處理，請考慮以下方法：

#### 使用提示快取

使用「prompt caching」（[提示快取](https://platform.claude.com/docs/zh-TW/build-with-claude/prompt-caching)）快取 PDF，以提升重複查詢的效能：

<CodeGroup>
  ```bash cURL
  curl -sL "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf" | base64 | tr -d '\n' > pdf_base64.txt
  # 使用 pdf_base64.txt 的內容建立 JSON 請求檔案
  jq -n --rawfile PDF_BASE64 pdf_base64.txt '{
      "model": "claude-opus-5-5",
      "max_tokens": 1024,
      "messages": [{
          "role": "user",
          "content": [{
              "type": "document",
              "source": {
                  "type": "base64",
                  "media_type": "application/pdf",
                  "data": $PDF_BASE64
              },
              "cache_control": {
                  "type": "ephemeral"
              }
          },
          {
              "type": "text",
              "text": "Which model has the highest human preference win rates across each use-case?"
          }]
      }]
  }' > request.json

  # 接著使用該 JSON 檔案呼叫 API
  curl https://api.anthropic.com/v1/messages \
    -H "content-type: application/json" \
    -H "x-api-key: $ANTHROPIC_API_KEY" \
    -H "anthropic-version: 2023-06-01" \
    -d @request.json
  ```

  ```bash CLI
  ant messages create --transform content --format yaml <<'YAML'
  model: claude-opus-5-5
  max_tokens: 1024
  messages:
    - role: user
      content:
        - type: document
          source:
            type: base64
            media_type: application/pdf
            data: "@./document.pdf"
          cache_control:
            type: ephemeral
        - type: text
          text: Which model has the highest human preference win rates across each use-case?
  YAML
  ```

  ```python Python
  import base64
  import httpx2

  # 首先，載入 PDF 並進行編碼
  pdf_url = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
  pdf_data = base64.standard_b64encode(
      httpx2.get(pdf_url, follow_redirects=True).content
  ).decode("utf-8")

  # 使用已快取的文件建立訊息
  client = anthropic.Anthropic()
  message = client.messages.create(
      model="claude-opus-5-5",
      max_tokens=1024,
      messages=[
          {
              "role": "user",
              "content": [
                  {
                      "type": "document",
                      "source": {
                          "type": "base64",
                          "media_type": "application/pdf",
                          "data": pdf_data,
                      },
                      "cache_control": {"type": "ephemeral"},
                  },
                  {
                      "type": "text",
                      "text": "Which model has the highest human preference win rates across each use-case?",
                  },
              ],
          }
      ],
  )

  print(message.content)
  ```

  ```typescript TypeScript
  // 首先，載入 PDF 並進行編碼
  const pdfURL =
    "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf";
  const pdfResponse = await fetch(pdfURL);
  const arrayBuffer = await pdfResponse.arrayBuffer();
  const pdfBase64 = Buffer.from(arrayBuffer).toString("base64");

  // 使用已快取的文件建立訊息
  const anthropic = new Anthropic();
  const response = await anthropic.messages.create({
    model: "claude-opus-5-5",
    max_tokens: 1024,
    messages: [
      {
        role: "user",
        content: [
          {
            type: "document",
            source: {
              type: "base64",
              media_type: "application/pdf",
              data: pdfBase64
            },
            cache_control: { type: "ephemeral" }
          },
          {
            type: "text",
            text: "Which model has the highest human preference win rates across each use-case?"
          }
        ]
      }
    ]
  });

  console.log(response);
  ```

  ```csharp C#
  var client = new AnthropicClient();

  // 下載並編碼 PDF
  var pdfUrl = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf";
  using var httpClient = new HttpClient();
  var pdfBase64 = Convert.ToBase64String(await httpClient.GetByteArrayAsync(pdfUrl));

  var message = await client.Messages.Create(new MessageCreateParams
  {
      Model = Model.ClaudeOpus5_5,
      MaxTokens = 1024,
      Messages =
      [
          new()
          {
              Role = Role.User,
              Content = new List<ContentBlockParam>
              {
                  new DocumentBlockParam
                  {
                      Source = new Base64PdfSource { Data = pdfBase64 },
                      CacheControl = new CacheControlEphemeral(),
                  },
                  new TextBlockParam("Which model has the highest human preference win rates across each use-case?"),
              },
          },
      ],
  });

  Console.WriteLine(message);
  ```

  ```go Go
  // 首先，載入 PDF 並進行編碼
  pdfURL := "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
  resp, err := http.Get(pdfURL)
  if err != nil {
  	panic(err)
  }
  defer resp.Body.Close()
  pdfBytes, err := io.ReadAll(resp.Body)
  if err != nil {
  	panic(err)
  }
  pdfBase64 := base64.StdEncoding.EncodeToString(pdfBytes)

  // 建立帶有快取控制的文件區塊
  client := anthropic.NewClient()
  message, err := client.Messages.New(context.TODO(), anthropic.MessageNewParams{
  	Model:     anthropic.ModelClaudeOpus5_5,
  	MaxTokens: 1024,
  	Messages: []anthropic.MessageParam{
  		anthropic.NewUserMessage(
  			anthropic.ContentBlockParamUnion{
  				OfDocument: &anthropic.DocumentBlockParam{
  					Source: anthropic.DocumentBlockParamSourceUnion{
  						OfBase64: &anthropic.Base64PDFSourceParam{
  							Data: pdfBase64,
  						},
  					},
  					CacheControl: anthropic.NewCacheControlEphemeralParam(),
  				},
  			},
  			anthropic.NewTextBlock("Which model has the highest human preference win rates across each use-case?"),
  		),
  	},
  })
  if err != nil {
  	panic(err)
  }

  fmt.Printf("%+v\n", message.Content)
  ```

  ```java Java
  AnthropicClient client = AnthropicOkHttpClient.fromEnv();

  // 下載 PDF 並進行編碼
  String pdfUrl =
    "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf";
  HttpClient httpClient = HttpClient.newBuilder().followRedirects(HttpClient.Redirect.NORMAL).build();
  HttpRequest request = HttpRequest.newBuilder().uri(URI.create(pdfUrl)).GET().build();

  HttpResponse<byte[]> response = httpClient.send(
    request,
    HttpResponse.BodyHandlers.ofByteArray()
  );
  String pdfBase64 = Base64.getEncoder().encodeToString(response.body());

  MessageCreateParams params = MessageCreateParams.builder()
    .model(Model.CLAUDE_OPUS_5_5)
    .maxTokens(1024)
    .addUserMessageOfBlockParams(
      List.of(
        ContentBlockParam.ofDocument(
          DocumentBlockParam.builder()
            .source(Base64PdfSource.builder().data(pdfBase64).build())
            .cacheControl(CacheControlEphemeral.builder().build())
            .build()
        ),
        ContentBlockParam.ofText(
          TextBlockParam.builder()
            .text(
              "Which model has the highest human preference win rates across each use-case?"
            )
            .build()
        )
      )
    )
    .build();

  Message message = client.messages().create(params);
  System.out.println(message);
  ```

  ```php PHP
  $client = new Client();

  // 載入 PDF 並進行編碼
  $pdf_url = 'https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf';
  $pdf_data = base64_encode(file_get_contents($pdf_url));

  $message = $client->messages->create(
      maxTokens: 1024,
      messages: [
          [
              'role' => 'user',
              'content' => [
                  [
                      'type' => 'document',
                      'source' => [
                          'type' => 'base64',
                          'media_type' => 'application/pdf',
                          'data' => $pdf_data,
                      ],
                      'cache_control' => ['type' => 'ephemeral'],
                  ],
                  [
                      'type' => 'text',
                      'text' => 'Which model has the highest human preference win rates across each use-case?',
                  ],
              ],
          ],
      ],
      model: 'claude-opus-5-5',
  );

  echo $message;
  ```

  ```ruby Ruby
  require "open-uri"

  # 載入 PDF 並進行編碼
  pdf_url = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
  pdf_bytes = URI.open(pdf_url, "rb") { |f| f.read }
  pdf_data = [pdf_bytes].pack("m0") # Base64-encode without newlines

  anthropic = Anthropic::Client.new

  message = anthropic.messages.create(
    model: "claude-opus-5-5",
    max_tokens: 1024,
    messages: [
      {
        role: "user",
        content: [
          {
            type: "document",
            source: {
              type: "base64",
              media_type: "application/pdf",
              data: pdf_data
            },
            cache_control: {type: "ephemeral"}
          },
          {
            type: "text",
            text: "Which model has the highest human preference win rates across each use-case?"
          }
        ]
      }
    ]
  )

  puts(message.content)
  ```
</CodeGroup>

#### 處理文件批次

使用 [Message Batches API](https://platform.claude.com/docs/zh-TW/build-with-claude/batch-processing) 在單一請求中處理多個 PDF：

<CodeGroup>
  ```bash cURL
  curl -sL "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf" | base64 | tr -d '\n' > pdf_base64.txt
  # 使用 pdf_base64.txt 的內容建立 JSON 請求檔案
  jq -n --rawfile PDF_BASE64 pdf_base64.txt '{
      "requests": [
      {
          "custom_id": "my-first-request",
          "params": {
              "model": "claude-opus-5-5",
              "max_tokens": 1024,
              "messages": [{
                  "role": "user",
                  "content": [{
                      "type": "document",
                      "source": {
                          "type": "base64",
                          "media_type": "application/pdf",
                          "data": $PDF_BASE64
                      }
                  },
                  {
                      "type": "text",
                      "text": "Which model has the highest human preference win rates across each use-case?"
                  }]
              }]
          }
      },
      {
          "custom_id": "my-second-request",
          "params": {
              "model": "claude-opus-5-5",
              "max_tokens": 1024,
              "messages": [{
                  "role": "user",
                  "content": [{
                      "type": "document",
                      "source": {
                          "type": "base64",
                          "media_type": "application/pdf",
                          "data": $PDF_BASE64
                      }
                  },
                  {
                      "type": "text",
                      "text": "Extract 5 key insights from this document."
                  }]
              }]
          }
      }]
  }' > request.json

  # 接著使用該 JSON 檔案呼叫 API
  curl https://api.anthropic.com/v1/messages/batches \
    -H "content-type: application/json" \
    -H "x-api-key: $ANTHROPIC_API_KEY" \
    -H "anthropic-version: 2023-06-01" \
    -d @request.json
  ```

  ```bash CLI
  ant messages:batches create <<'YAML'
  requests:
    - custom_id: my-first-request
      params:
        model: claude-opus-5-5
        max_tokens: 1024
        messages:
          - role: user
            content:
              - type: document
                source:
                  type: base64
                  media_type: application/pdf
                  data: "@./document.pdf"
              - type: text
                text: >-
                  Which model has the highest human preference win rates
                  across each use-case?
    - custom_id: my-second-request
      params:
        model: claude-opus-5-5
        max_tokens: 1024
        messages:
          - role: user
            content:
              - type: document
                source:
                  type: base64
                  media_type: application/pdf
                  data: "@./document.pdf"
              - type: text
                text: Extract 5 key insights from this document.
  YAML
  ```

  ```python Python
  import base64
  import httpx2

  # 首先，載入 PDF 並進行編碼
  pdf_url = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
  pdf_data = base64.standard_b64encode(
      httpx2.get(pdf_url, follow_redirects=True).content
  ).decode("utf-8")

  # 建立一批使用該文件的請求
  client = anthropic.Anthropic()
  message_batch = client.messages.batches.create(
      requests=[
          {
              "custom_id": "my-first-request",
              "params": {
                  "model": "claude-opus-5-5",
                  "max_tokens": 1024,
                  "messages": [
                      {
                          "role": "user",
                          "content": [
                              {
                                  "type": "document",
                                  "source": {
                                      "type": "base64",
                                      "media_type": "application/pdf",
                                      "data": pdf_data,
                                  },
                              },
                              {
                                  "type": "text",
                                  "text": "Which model has the highest human preference win rates across each use-case?",
                              },
                          ],
                      }
                  ],
              },
          },
          {
              "custom_id": "my-second-request",
              "params": {
                  "model": "claude-opus-5-5",
                  "max_tokens": 1024,
                  "messages": [
                      {
                          "role": "user",
                          "content": [
                              {
                                  "type": "document",
                                  "source": {
                                      "type": "base64",
                                      "media_type": "application/pdf",
                                      "data": pdf_data,
                                  },
                              },
                              {
                                  "type": "text",
                                  "text": "Extract 5 key insights from this document.",
                              },
                          ],
                      }
                  ],
              },
          },
      ]
  )

  print(message_batch)
  ```

  ```typescript TypeScript
  // 首先，載入 PDF 並進行編碼
  const pdfURL =
    "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf";
  const pdfResponse = await fetch(pdfURL);
  const arrayBuffer = await pdfResponse.arrayBuffer();
  const pdfBase64 = Buffer.from(arrayBuffer).toString("base64");

  // 建立一批使用此文件的請求
  const anthropic = new Anthropic();
  const response = await anthropic.messages.batches.create({
    requests: [
      {
        custom_id: "my-first-request",
        params: {
          model: "claude-opus-5-5",
          max_tokens: 1024,
          messages: [
            {
              role: "user",
              content: [
                {
                  type: "document",
                  source: {
                    type: "base64",
                    media_type: "application/pdf",
                    data: pdfBase64
                  }
                },
                {
                  type: "text",
                  text: "Which model has the highest human preference win rates across each use-case?"
                }
              ]
            }
          ]
        }
      },
      {
        custom_id: "my-second-request",
        params: {
          model: "claude-opus-5-5",
          max_tokens: 1024,
          messages: [
            {
              role: "user",
              content: [
                {
                  type: "document",
                  source: {
                    type: "base64",
                    media_type: "application/pdf",
                    data: pdfBase64
                  }
                },
                {
                  type: "text",
                  text: "Extract 5 key insights from this document."
                }
              ]
            }
          ]
        }
      }
    ]
  });

  console.log(response);
  ```

  ```csharp C#
  var client = new AnthropicClient();

  // 下載 PDF 並進行編碼
  var pdfUrl = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf";
  using var httpClient = new HttpClient();
  var pdfBase64 = Convert.ToBase64String(await httpClient.GetByteArrayAsync(pdfUrl));

  var batch = await client.Messages.Batches.Create(new BatchCreateParams
  {
      Requests =
      [
          new()
          {
              CustomID = "my-first-request",
              Params = new()
              {
                  Model = Model.ClaudeOpus5_5,
                  MaxTokens = 1024,
                  Messages =
                  [
                      new()
                      {
                          Role = Role.User,
                          Content = new List<ContentBlockParam>
                          {
                              new DocumentBlockParam
                              {
                                  Source = new Base64PdfSource { Data = pdfBase64 },
                              },
                              new TextBlockParam("Which model has the highest human preference win rates across each use-case?"),
                          },
                      },
                  ],
              },
          },
          new()
          {
              CustomID = "my-second-request",
              Params = new()
              {
                  Model = Model.ClaudeOpus5_5,
                  MaxTokens = 1024,
                  Messages =
                  [
                      new()
                      {
                          Role = Role.User,
                          Content = new List<ContentBlockParam>
                          {
                              new DocumentBlockParam
                              {
                                  Source = new Base64PdfSource { Data = pdfBase64 },
                              },
                              new TextBlockParam("Extract 5 key insights from this document."),
                          },
                      },
                  ],
              },
          },
      ],
  });

  Console.WriteLine(batch);
  ```

  ```go Go
  // 首先，載入 PDF 並進行編碼
  pdfURL := "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
  resp, err := http.Get(pdfURL)
  if err != nil {
  	panic(err)
  }
  defer resp.Body.Close()
  pdfBytes, err := io.ReadAll(resp.Body)
  if err != nil {
  	panic(err)
  }
  pdfBase64 := base64.StdEncoding.EncodeToString(pdfBytes)

  // 建立一批使用此文件的請求
  client := anthropic.NewClient()
  batch, err := client.Messages.Batches.New(context.TODO(), anthropic.MessageBatchNewParams{
  	Requests: []anthropic.MessageBatchNewParamsRequest{
  		{
  			CustomID: "my-first-request",
  			Params: anthropic.MessageBatchNewParamsRequestParams{
  				Model:     anthropic.ModelClaudeOpus5_5,
  				MaxTokens: 1024,
  				Messages: []anthropic.MessageParam{
  					anthropic.NewUserMessage(
  						anthropic.NewDocumentBlock(anthropic.Base64PDFSourceParam{
  							Data: pdfBase64,
  						}),
  						anthropic.NewTextBlock("Which model has the highest human preference win rates across each use-case?"),
  					),
  				},
  			},
  		},
  		{
  			CustomID: "my-second-request",
  			Params: anthropic.MessageBatchNewParamsRequestParams{
  				Model:     anthropic.ModelClaudeOpus5_5,
  				MaxTokens: 1024,
  				Messages: []anthropic.MessageParam{
  					anthropic.NewUserMessage(
  						anthropic.NewDocumentBlock(anthropic.Base64PDFSourceParam{
  							Data: pdfBase64,
  						}),
  						anthropic.NewTextBlock("Extract 5 key insights from this document."),
  					),
  				},
  			},
  		},
  	},
  })
  if err != nil {
  	panic(err)
  }

  fmt.Printf("%+v\n", batch)
  ```

  ```java Java
  AnthropicClient client = AnthropicOkHttpClient.fromEnv();

  // 下載 PDF 並進行編碼
  String pdfUrl =
    "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf";
  HttpClient httpClient = HttpClient.newBuilder().followRedirects(HttpClient.Redirect.NORMAL).build();
  HttpRequest request = HttpRequest.newBuilder().uri(URI.create(pdfUrl)).GET().build();

  HttpResponse<byte[]> response = httpClient.send(
    request,
    HttpResponse.BodyHandlers.ofByteArray()
  );
  String pdfBase64 = Base64.getEncoder().encodeToString(response.body());

  BatchCreateParams params = BatchCreateParams.builder()
    .addRequest(
      BatchCreateParams.Request.builder()
        .customId("my-first-request")
        .params(
          BatchCreateParams.Request.Params.builder()
            .model(Model.CLAUDE_OPUS_5_5)
            .maxTokens(1024)
            .addUserMessageOfBlockParams(
              List.of(
                ContentBlockParam.ofDocument(
                  DocumentBlockParam.builder()
                    .source(Base64PdfSource.builder().data(pdfBase64).build())
                    .build()
                ),
                ContentBlockParam.ofText(
                  TextBlockParam.builder()
                    .text(
                      "Which model has the highest human preference win rates across each use-case?"
                    )
                    .build()
                )
              )
            )
            .build()
        )
        .build()
    )
    .addRequest(
      BatchCreateParams.Request.builder()
        .customId("my-second-request")
        .params(
          BatchCreateParams.Request.Params.builder()
            .model(Model.CLAUDE_OPUS_5_5)
            .maxTokens(1024)
            .addUserMessageOfBlockParams(
              List.of(
                ContentBlockParam.ofDocument(
                  DocumentBlockParam.builder()
                    .source(Base64PdfSource.builder().data(pdfBase64).build())
                    .build()
                ),
                ContentBlockParam.ofText(
                  TextBlockParam.builder()
                    .text("Extract 5 key insights from this document.")
                    .build()
                )
              )
            )
            .build()
        )
        .build()
    )
    .build();

  MessageBatch batch = client.messages().batches().create(params);
  System.out.println(batch);
  ```

  ```php PHP
  $client = new Client();

  // 載入 PDF 並進行編碼
  $pdf_url = 'https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf';
  $pdf_data = base64_encode(file_get_contents($pdf_url));

  $batch = $client->messages->batches->create(
      requests: [
          [
              'custom_id' => 'my-first-request',
              'params' => [
                  'model' => 'claude-opus-5-5',
                  'max_tokens' => 1024,
                  'messages' => [
                      [
                          'role' => 'user',
                          'content' => [
                              [
                                  'type' => 'document',
                                  'source' => [
                                      'type' => 'base64',
                                      'media_type' => 'application/pdf',
                                      'data' => $pdf_data,
                                  ],
                              ],
                              [
                                  'type' => 'text',
                                  'text' => 'Which model has the highest human preference win rates across each use-case?',
                              ],
                          ],
                      ],
                  ],
              ],
          ],
          [
              'custom_id' => 'my-second-request',
              'params' => [
                  'model' => 'claude-opus-5-5',
                  'max_tokens' => 1024,
                  'messages' => [
                      [
                          'role' => 'user',
                          'content' => [
                              [
                                  'type' => 'document',
                                  'source' => [
                                      'type' => 'base64',
                                      'media_type' => 'application/pdf',
                                      'data' => $pdf_data,
                                  ],
                              ],
                              [
                                  'type' => 'text',
                                  'text' => 'Extract 5 key insights from this document.',
                              ],
                          ],
                      ],
                  ],
              ],
          ],
      ],
  );

  echo $batch;
  ```

  ```ruby Ruby
  require "open-uri"

  # 載入 PDF 並進行編碼
  pdf_url = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
  pdf_bytes = URI.open(pdf_url, "rb") { |f| f.read }
  pdf_data = [pdf_bytes].pack("m0") # Base64-encode without newlines

  anthropic = Anthropic::Client.new

  message_batch = anthropic.messages.batches.create(
    requests: [
      {
        custom_id: "my-first-request",
        params: {
          model: "claude-opus-5-5",
          max_tokens: 1024,
          messages: [
            {
              role: "user",
              content: [
                {
                  type: "document",
                  source: {
                    type: "base64",
                    media_type: "application/pdf",
                    data: pdf_data
                  }
                },
                {
                  type: "text",
                  text: "Which model has the highest human preference win rates across each use-case?"
                }
              ]
            }
          ]
        }
      },
      {
        custom_id: "my-second-request",
        params: {
          model: "claude-opus-5-5",
          max_tokens: 1024,
          messages: [
            {
              role: "user",
              content: [
                {
                  type: "document",
                  source: {
                    type: "base64",
                    media_type: "application/pdf",
                    data: pdf_data
                  }
                },
                {
                  type: "text",
                  text: "Extract 5 key insights from this document."
                }
              ]
            }
          ]
        }
      }
    ]
  )

  puts(message_batch)
  ```
</CodeGroup>

批次以非同步方式處理。若要檢查進度並在處理結束後擷取結果，請參閱[批次處理](https://platform.claude.com/docs/zh-TW/build-with-claude/batch-processing)。

## 後續步驟

<CardGroup cols={2}>
  <Card title="視覺" icon="image" href="https://platform.claude.com/docs/zh-TW/build-with-claude/vision">
    Claude 的視覺能力使其能夠理解並分析影像，為多模態互動開啟令人興奮的可能性。
  </Card>

  <Card title="試用 PDF 範例" icon="file" href="https://platform.claude.com/cookbook/multimodal-getting-started-with-vision">
    在 Claude Cookbook 食譜中探索 PDF 處理的實用範例。
  </Card>

  <Card title="檢視 API 參考" icon="code" href="https://platform.claude.com/docs/zh-TW/api/messages/create">
    查看 PDF 支援的完整 API 文件。
  </Card>
</CardGroup>