从肉眼筛查到标准化校验：三大主流表格解析方案实测

一、引言：复杂表格解析的隐形难题

在处理大批量文档中的表格数据时，我们团队长期被低效的解析质量验证方式和参差不齐的解析工具所困扰——接入过通用OCR API，也测试过多款开源解析方案，却始终没能解决复杂表格（多层表头、合并单元格、跨页续表、嵌套子表）的解析准确性问题。随机抽检的“肉眼筛查”模式，既无法覆盖所有风险，又极易让结构错位、字段归属错误的表格数据流入RAG系统、知识库或风控模型，导致下游业务决策失误。

为了评估当前主流表格解析方案在真实生产场景中的表现，我们选取了市面上关注度较高的几款工具进行横向测试，包括合合信息推出的 TextIn xParse、PaddleOCR、MinerU 等。

本次测评将从结构、关系、内容准等多个维度展开，并结合实际复杂表格案例，观察不同方案在解析效果和可用性上的差异。 究竟哪种方案更适合落地到企业级应用场景？接下来我们直接看结果。

二、三维度自查清单：系统性地判断解析结果是否可用

这套检查方法将表格解析质量分为三个层次，检查成本依次递增：

自查策略： 先看结构（一眼可见），再看关系（需要对照原文），最后抽查内容（高风险区域抽样）。

2.1 维度一：结构对不对

结构是表格解析的第一关。如果结构层面出了问题，后续的关系归属和内容精度就没有讨论的意义。

结构自查围绕四个问题：

🔬 实测案例1：复杂合并单元格工程表

业务场景： 某基坑支护设计规范表，表格同时存在大面积行合并、列合并和跨层级表头。单个单元格内容往往需要关联多行、多列信息才能完整理解其含义。对于表格解析系统而言，不仅要恢复表格结构，更要准确保留“结构类型—安全等级—适用条件”之间的层级关系，否则容易出现字段挂错、内容归属错误等问题。

原表特征：

• Paddle OCR:

测试结果：表格区域错误，合并单元格识别有误

出现行列合并关系丢失，原图是高度结构化的多层嵌套表格（有大合并单元格、不同层级的子表格），识别后变成了松散的文本块；其次表头混乱，原图最上方的“结构类型”“安全等级”“适用条件”等表头基本丢失或错位。

• MinerU:

测试结果：整体表格结构识别正确，多级表头、合并单元格及字段归属关系均得到较好保留，能够还原原始表格的层级结构。

但在细节识别上仍存在一定问题，右侧序号1-5每行的文字推挤成一块，可读性较弱。

• TextIn xParse

测试结果：整体表格结构识别正确，多级表头、合并单元格及字段归属关系均得到较好保留，能够还原原始表格的层级结构。

细节部分出现数字识别错误，对比MinerU对右侧大段内容内容的识别情况，每行显示清晰。

2.2 维度二：关系对不对

关系层比结构层更隐蔽，但直接影响数据的业务含义。你拿到的数据可能行列数量都对、合并单元格也在，但某个数字挂到了错误的表头上——这比结构错误更难发现，危害也更大。

关系自查的核心问题：

🔬 实测案例：嵌套多个表的劳务报酬审批单

业务场景： 某企业员工竞业补贴审批单，主表中嵌套了两张独立的追踪表，分别用于记录2025年和2026年的补贴回收情况。两张追踪表均包含独立标题、月份维度及金额数据，需要准确识别表格边界、层级关系以及金额对应月份，避免出现表格合并或数据错位的问题。

原表特征：

实测对比：

• MinerU：

测试结果：嵌套关系丢失

原文中包含两个独立的嵌套表：“2025年收回竞业补贴追踪表”和“2026年收回竞业补贴追踪表”，两张表均有各自的标题、表头和数据区域。

但在解析结果中，系统未能识别两张表之间的边界，将两个嵌套表直接合并成了一张连续表格。其中，第一张表的标题完全丢失，第二张表的标题则被当作普通内容插入到了表格中间。

• PaddleOCR：

测试结果：嵌套关系丢失

同上，原文中包含两个独立的嵌套表，但解析结果未能识别表格边界，将两张表直接拼接在一起，导致嵌套结构丢失；原始标题也未能识别。

此外，表格中的数据归属出现错误。原本应位于11月份列下的金额“200”，被错误识别到了10月份列，导致数据位置发生偏移。

这类错误看似只是列位置偏移，但实际会直接影响数据准确性。尤其是在财务、审计、经营分析等场景中，金额与时间维度的对应关系至关重要，一旦月份与金额错位，可能导致统计结果失真、趋势判断错误，甚至影响后续业务决策。

• TextIn xParse：

测试结果：识别正确

使用TextIn Xparse的vlm解析模式，两个嵌套表均被正确识别和拆分，标题、表头及数据位置保持一致，金额与月份对应关系准确，较好地还原了原始表格结构。

2.3 维度三：内容对不对

如果结构和关系都已通过检查，内容层通常属于低概率出错区。但某些高风险区域仍需重点抽查。

内容自查重点关注：

🔬 实测案例：审计底稿中的密集数字表格

业务场景： 随便绘制的AI使用量统计表，包含大量金额数字、百分比、以及脚注小字。表格为扫描件，底部备注使用 7pt 小号字体，且页面有浅色水印覆盖。

原表特征：

文字密集，有些文字模糊不清，有盖章，有签字，有单元格合并，批注等。

注：下图数据仅是随机填写的测试数据，纯属虚构，如有雷同，联系博主

对比差异：

xParse表现稳定，paddle_ocr和mineru少量数字千分号丢失错误，少量运算符号识别丢失错误。

结论： 内容层的错误往往是"细节杀手"——一个丢失的负号意味着坏账变成了收益，一个丢失的百分号意味着比例变成了整数。xParse 在密集数字和低对比度小字两个高风险区域的表现稳定，尤其是对扫描件中小号字体的处理明显优于对比方案。

三、TextIn xParse 产品初体验

3.1 产品定位

TextIn xParse 是合合信息旗下面向 LLM 与 Agent 的文档智能基础设施。它的核心能力是将复杂文档（PDF、扫描件、图片）转化为干净、一致、可用于 AI 的结构化数据，适用于知识库构建、RAG 系统、或其他 AI 工作流的下游应用。

与通用 OCR 工具不同，xParse 专门针对复杂表格做了深度优化——不只是"识别文字"，而是理解表格的结构和语义。

3.2 产品体验流程

访问 https://www.textin.com/register/code/VT4KAZ，登录后，可以直接体验表格解析的全流程。整体操作路径非常简洁，操作流程共三步：

第一步：上传文档。 支持 PDF、PNG、JPG 等格式，可直接拖拽或点击上传。我们选择一张包含复杂表格的 PDF 页面进行测试。

截图描述： 展示文件上传区域，支持拖拽上传或本地文件选择。上传区域下方显示支持的文件格式列表（PDF/PNG/JPG），以及单文件大小限制说明。已上传的文件以缩略图形式显示在左侧列表中。

第二步：自动解析。 xParse 提供多种解析模式，针对表格场景可选择"表格解析"模式。

第三步：查看解析结果。 点击"开始解析"后，右侧面板实时渲染解析结果——包括 JSON 结构化数据和可视化表格渲染等多种解析结果。这个"原文 ↔ 解析结果"并排展示的方式，让比对变得直观。

第四步：可以导出常用的格式。解析后一键导出MD、txt、excel、word，格式转换没烦恼！

第五步：加入知识库。准确快速回答你想知道的相关数据信息！

体验小结： 从上传到拿到结构化结果，整个过程不超过 30 秒。界面的"原文 ↔ 解析结果"同屏对比设计在实际使用中非常实用，不需要在多个窗口之间来回切换。

四、python调用api解析文档，三步搞定

第一步：获取API KEY

使用文档解析API处理文档时，您需要先获取API Key。请先登录后前往 TextIn工作台 – 账号与开发者信息 获取您的 x-ti-app-id 和 x-ti-secret-code 。点击获取：https://www.textin.com/console/dashboard/setting

第二步：代码里面改上自己的 x-ti-app-id 和 x-ti-secret-code

import json
import requests
from typing import Dict, Optional
from dataclasses import dataclass

class OCRClient:
def __init__(self, app_id: str, secret_code: str):
self.app_id = app_id
self.secret_code = secret_code

def recognize(self, file_content: bytes, options: dict) -> str:
# 构建请求参数
params = {}
for key, value in options.items():
params[key] = str(value)

# 设置请求头
headers = {
"x-ti-app-id": self.app_id,
"x-ti-secret-code": self.secret_code,
"x-ti-client-source": "sample-code-v1.0",
# 方式一：读取本地文件
"Content-Type": "application/octet-stream"
# 方式二：使用URL方式
# "Content-Type": "text/plain"
}

# 发送请求
response = requests.post(
f"https://api.textin.com/ai/service/v1/pdf_to_markdown",
params=params,
headers=headers,
data=file_content
)

# 检查响应状态
response.raise_for_status()
return response.text

def main():
# 创建客户端实例
client = OCRClient("自己的x-ti-app-id", "自己的x-ti-secret-code")

# 读取图片文件
# 方式一：读取本地文件
with open("sample_02.pdf", "rb") as f:
file_content = f.read()
# 方式二：使用URL方式（需要将headers中的Content-Type改为'text/plain'）
# file_content = "https://example.com/path/to/your.pdf"

# 设置转换选项
options = dict(
apply_document_tree=1,
apply_merge=1,
catalog_details=1,
dpi=144,
formula_level=1,
get_excel=1,
get_image="objects",
markdown_details=1,
page_count=1000,
page_details=1,
paratext_mode="annotation",
parse_mode="scan",
table_flavor="html",
)

try:
response = client.recognize(file_content, options)

# 保存完整的JSON响应到result.json文件
with open("result.json", "w", encoding="utf-8") as f:
f.write(response)

# 解析JSON响应以提取markdown内容
json_response = json.loads(response)
if "result" in json_response and "markdown" in json_response["result"]:
markdown_content = json_response["result"]["markdown"]
with open("result.md", "w", encoding="utf-8") as f:
f.write(markdown_content)

print("Success! Result saved to result.json and result.md")
except Exception as e:
print(f"Error: {e}")

if __name__ == "__main__":
main()

第三步： 解析自己的文件，我这里放的是sample_02.pdf文件，你可以自定义，代码注释详细，可以自定义配置。

一图了解整个流程

五、从解析到应用：下游集成实战

拿到解析结果之后，开发者最关心的问题是：怎么把它接入我的 RAG 系统或知识库？

以 TextIn xParse 的 JSON 输出为例，其结构化字段设计充分考虑了 AI 工作流的下游需求：

{"elements":[{"coordinates":[0.033155,0.010526,0.305348,0.010526,0.305348,0.014899,0.033155,0.014899],"element_id":"b0b28b7598462b1dc35628df9bd3086dbdfa0c4f15fea70bb9a4e328f07e137a","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"一、各部门Token用量与费用明细 核心报表","type":"Header"},{"coordinates":[0.031016,0.025749,0.960963,0.025101,0.962032,0.28,0.031551,0.28081],"element_id":"005d44743d4991491f5d4bada074ab5f22e4466730fc16d1bf18d30f8551f7d3","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"ref_element_id":"f7fe5462db30f641fafe7ab4f6c144c9310937325cb997d94fc507ca257ec08e"},"page_number":1,"sub_type":"borderless","text":"<table border=\\"1\\" ><tr>\\n<td rowspan=\\"2\\">部门</td>\\n<td rowspan=\\"2\\">二级部门</td>\\n<td rowspan=\\"2\\">负责人</td>\\n<td rowspan=\\"2\\">人数</td>\\n<td rowspan=\\"2\\">主要模型</td>\\n<td colspan=\\"2\\">Token用量</td>\\n<td rowspan=\\"2\\">费用（¥）</td>\\n<td rowspan=\\"2\\">预算（¥）</td>\\n<td rowspan=\\"2\\">执行率</td>\\n</tr><tr>\\n<td>输入Token</td>\\n<td>输出Token</td>\\n</tr><tr>\\n<td rowspan=\\"4\\">研发部</td>\\n<td>合计·95人</td>\\n<td></td>\\n<td>95</td>\\n<td>-</td>\\n<td>2,033.40亿</td>\\n<td>251.00亿</td>\\n<td>6,535,789.50</td>\\n<td>7,200,000.00</td>\\n<td>90.8%</td>\\n</tr><tr>\\n<td>AI平台组</td>\\n<td>张明远</td>\\n<td>45</td>\\n<td>GPT-4/Claude Opus／文心一言4.0</td>\\n<td>1,796.00亿</td>\\n<td>251.00亿</td>\\n<td>6,205,789.50</td>\\n<td>6,800,000.00</td>\\n<td>91.3%</td>\\n</tr><tr>\\n<td>基础架构组</td>\\n<td>李雪琴</td>\\n<td>32</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>770.00亿</td>\\n<td>115.00亿</td>\\n<td>2,670,000.00</td>\\n<td>3,000,000.00</td>\\n<td>89.0%</td>\\n</tr><tr>\\n<td>测试效能组</td>\\n<td>王建国</td>\\n<td>18</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>198.00亿</td>\\n<td>37.00亿</td>\\n<td>690,000.00</td>\\n<td>800,000.00</td>\\n<td>86.3%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">产品部</td>\\n<td>合计·50人</td>\\n<td></td>\\n<td>50</td>\\n<td>-</td>\\n<td>1,780.00亿</td>\\n<td>240.00亿</td>\\n<td>5,900,000.00</td>\\n<td>6,500,000.00</td>\\n<td>90.8%</td>\\n</tr><tr>\\n<td>智能产品线</td>\\n<td>赵雨婷</td>\\n<td>28</td>\\n<td>GPT-4/Claude Opus／通义千问</td>\\n<td>1,150.00亿</td>\\n<td>195.00亿</td>\\n<td>4,240,000.00</td>\\n<td>4,600,000.00</td>\\n<td>92.2%</td>\\n</tr><tr>\\n<td>数据产品线</td>\\n<td>孙浩然</td>\\n<td>22</td>\\n<td>GPT-4 / Claude Sonnet／文心一言4.0</td>\\n<td>740.00亿</td>\\n<td>115.00亿</td>\\n<td>2,330,000.00</td>\\n<td>2,600,000.00</td>\\n<td>89.6%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">市场部</td>\\n<td>合计·50人</td>\\n<td></td>\\n<td>50</td>\\n<td>-</td>\\n<td>1,195.00亿</td>\\n<td>206.00亿</td>\\n<td>5,453,000.00</td>\\n<td>5,800,000.00</td>\\n<td>94.0%</td>\\n</tr><tr>\\n<td>内容营销组</td>\\n<td>周思远</td>\\n<td>35</td>\\n<td>GPT-4/ Claude Opus/Claude Sonnet</td>\\n<td>950.00亿</td>\\n<td>257.00亿</td>\\n<td>5,450,000.00</td>\\n<td>5,800,000.00</td>\\n<td>94.0%</td>\\n</tr><tr>\\n<td>品牌设计组</td>\\n<td>吴佳琪</td>\\n<td>15</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>185.00亿</td>\\n<td>84.00亿</td>\\n<td>1,023,000.00</td>\\n<td>1,200,000.00</td>\\n<td>85.3%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">运营部</td>\\n<td>合计·65人</td>\\n<td></td>\\n<td>65</td>\\n<td>-</td>\\n<td>1,990.00亿</td>\\n<td>485.00亿</td>\\n<td>10,800,000.00</td>\\n<td>12,000,000.00</td>\\n<td>90.0%</td>\\n</tr><tr>\\n<td>用户运营组</td>\\n<td>郑子涵</td>\\n<td>40</td>\\n<td>GPT-4/Claude Opus/Claude Sonnet／文心一言4.0</td>\\n<td>1,990.00亿</td>\\n<td>785.00亿</td>\\n<td>10,550,000.00</td>\\n<td>11,500,000.00</td>\\n<td>91.7%</td>\\n</tr><tr>\\n<td>数据运营组</td>\\n<td>陆雨萱</td>\\n<td>25</td>\\n<td>GPT-4/Claude Opus／通义千问</td>\\n<td>880.00亿</td>\\n<td>425.00亿</td>\\n<td>5,010,000.00</td>\\n<td>5,500,000.00</td>\\n<td>91.1%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">数据科学部</td>\\n<td>合计·58人</td>\\n<td></td>\\n<td>58</td>\\n<td>-</td>\\n<td>2,940.00亿</td>\\n<td>372.00亿</td>\\n<td>14,814,000.00</td>\\n<td>16,000,000.00</td>\\n<td>92.6%</td>\\n</tr><tr>\\n<td>算法研究组</td>\\n<td>钱伟豪</td>\\n<td>38</td>\\n<td>GPT-4 / Claude Opus / Claude Sonnet／文心一言4.0／通义千问</td>\\n<td>2,950.00亿</td>\\n<td>467.00亿</td>\\n<td>14,810,000.00</td>\\n<td>16,000,000.00</td>\\n<td>92.6%</td>\\n</tr><tr>\\n<td>数据工程组</td>\\n<td>陈俊杰</td>\\n<td>20</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>248.00亿</td>\\n<td>70.00亿</td>\\n<td>1,134,000.00</td>\\n<td>1,300,000.00</td>\\n<td>87.2%</td>\\n</tr><tr>\\n<td rowspan=\\"2\\">财务部</td>\\n<td>合计·12人</td>\\n<td></td>\\n<td>12</td>\\n<td>-</td>\\n<td>90.00亿</td>\\n<td>27.50亿</td>\\n<td>434,000.00</td>\\n<td>500,000.00</td>\\n<td>86.8%</td>\\n</tr><tr>\\n<td>财务分析组</td>\\n<td>林妙音</td>\\n<td>12</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>90.00亿</td>\\n<td>27.50亿</td>\\n<td>434,000.00</td>\\n<td>500,000.00</td>\\n<td>86.8%</td>\\n</tr><tr>\\n<td>集团合计</td>\\n<td></td>\\n<td></td>\\n<td></td>\\n<td>-</td>\\n<td>10,028.40亿</td>\\n<td>1,906.50亿</td>\\n<td>¥43,936,789.50</td>\\n<td>¥48,000,000.00</td>\\n<td>91.5%</td>\\n</tr></table>","type":"Table"},{"coordinates":[0.034225,0.300567,0.213369,0.300567,0.213369,0.304939,0.034225,0.304939],"element_id":"f7fe5462db30f641fafe7ab4f6c144c9310937325cb997d94fc507ca257ec08e","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"**二、各模型用量与费用对比**","type":"TableCaption"},{"coordinates":[0.031551,0.313684,0.5,0.313684,0.49893,0.410364,0.029412,0.410364],"element_id":"913ee7e62fc92de95e4d7c80b9a84ded6ef77ceb22c86649a7f556f8cad5dd45","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"ref_element_id":"74c3526f3d69b8129067aaf8df191f4fdba8d7e9bab184723712ce93b7ed92a1"},"page_number":1,"sub_type":"borderless","text":"<table border=\\"1\\" ><tr>\\n<td>模型</td>\\n<td>调用次数</td>\\n<td>Token小计</td>\\n<td>总费用（¥）</td>\\n<td>干Token均价（¥）</td>\\n<td>费用占比</td>\\n</tr><tr>\\n<td>GPT-4</td>\\n<td>15,680,000</td>\\n<td>4,720.00亿</td>\\n<td>21,278,000.00</td>\\n<td>0.045</td>\\n<td>43.2%</td>\\n</tr><tr>\\n<td>ClaudeOpus</td>\\n<td>8,950,000</td>\\n<td>3,515.50亿</td>\\n<td>15,645,000.00</td>\\n<td>0.044</td>\\n<td>31.8%</td>\\n</tr><tr>\\n<td>ClaudeSonnet</td>\\n<td>8,720,000</td>\\n<td>2,034.00亿</td>\\n<td>6,885,000.00</td>\\n<td>0.034</td>\\n<td>14.0%</td>\\n</tr><tr>\\n<td>文心一言4.0</td>\\n<td>4,670,000</td>\\n<td>1,267.20亿</td>\\n<td>3,310,000.00</td>\\n<td>0.026</td>\\n<td>6.7%</td>\\n</tr><tr>\\n<td>通义千问</td>\\n<td>2,960,000</td>\\n<td>724.50亿</td>\\n<td>1,570,000.00</td>\\n<td>0.022</td>\\n<td>3.2%</td>\\n</tr><tr>\\n<td>合计</td>\\n<td>40,980,000</td>\\n<td>12,261.20亿</td>\\n<td>48,688,000.00</td>\\n<td>-</td>\\n<td>100%</td>\\n</tr></table>","type":"Table"},{"coordinates":[0.568449,0.300567,0.728877,0.300567,0.728877,0.305101,0.568449,0.305101],"element_id":"74c3526f3d69b8129067aaf8df191f4fdba8d7e9bab184723712ce93b7ed92a1","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"**三、月度趋势（2025年）**","type":"TableCaption"},{"coordinates":[0.565775,0.313036,0.962567,0.313198,0.963102,0.443725,0.565775,0.443725],"element_id":"151a7f0d9eecc434cbe14a1b79375060f837ec40226c689e12e885d3692ebeba","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"sub_type":"borderless","text":"<table border=\\"1\\" ><tr>\\n<td>月份</td>\\n<td>调用次数</td>\\n<td>Token总量</td>\\n<td>费用（¥）</td>\\n<td>环比（费用）</td>\\n</tr><tr>\\n<td>1月</td>\\n<td>2,575,000</td>\\n<td>485.00亿</td>\\n<td>2,722,000.00</td>\\n<td>-</td>\\n</tr><tr>\\n<td>2月</td>\\n<td>2,472,000</td>\\n<td>464.00亿</td>\\n<td>2,605,000.00</td>\\n<td>-4.3%<img src=\\"https://web-api.textin.com/ocr_image/external/2a10b2c1397dab4b.jpg\\"></td>\\n</tr><tr>\\n<td>3月</td>\\n<td>3,038,000</td>\\n<td>572.00亿</td>\\n<td>3,205,000.00</td>\\n<td>+23.0%</td>\\n</tr><tr>\\n<td>4月</td>\\n<td>3,263,000</td>\\n<td>614.00亿</td>\\n<td>3,446,000.00</td>\\n<td>+7.5%</td>\\n</tr><tr>\\n<td>5月</td>\\n<td>3,464,000</td>\\n<td>652.00亿</td>\\n<td>3,659,000.00</td>\\n<td>+6.2%</td>\\n</tr><tr>\\n<td>6月</td>\\n<td>3,630,000</td>\\n<td>683.00亿</td>\\n<td>3,839,000.00</td>\\n<td>+4.9%</td>\\n</tr><tr>\\n<td>7月</td>\\n<td>3,538,000</td>\\n<td>665.00亿</td>\\n<td>3,733,000.00</td>\\n<td>-2.8%</td>\\n</tr><tr>\\n<td>8月</td>\\n<td>3,721,000</td>\\n<td>700.00亿</td>\\n<td>3,936,000.00</td>\\n<td>+5.4%</td>\\n</tr><tr>\\n<td>9月</td>\\n<td>3,646,000</td>\\n<td>685.00亿</td>\\n<td>3,850,000.00</td>\\n<td>-2.2%</td>\\n</tr><tr>\\n<td>10月</td>\\n<td>3,839,000</td>\\n<td>721.00亿</td>\\n<td>4,054,000.00</td>\\n<td>+5.3%</td>\\n</tr><tr>\\n<td>11月</td>\\n<td>3,928,000</td>\\n<td>739.00亿</td>\\n<td>4,153,000.00</td>\\n<td>+2.4%</td>\\n</tr><tr>\\n<td>12月</td>\\n<td>4,146,000</td>\\n<td>780.00亿</td>\\n<td>4,388,000.00</td>\\n<td>+5.7%</td>\\n</tr><tr>\\n<td>累计</td>\\n<td>42,260,000</td>\\n<td>7,760.00亿</td>\\n<td>43,590,000.00</td>\\n<td>-</td>\\n</tr></table>","type":"Table"},{"coordinates":[0.568984,0.446802,0.948663,0.447449,0.948663,0.495223,0.568984,0.494575],"element_id":"3718ef78786ac3fe0afc9a7ea16d69c0688a0739cfa32f5448e8f0dfa32e295b","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":299,"is_continuation":false,"width":710},"page_number":1,"sub_type":"chart","text":"440385月度费用366384373394345321400272261(A)350旺邮3002501月2月3月4月5月6月7月8月9月10月11月12月","type":"Image"},{"coordinates":[0.03369,0.521943,0.201604,0.521943,0.201604,0.52664,0.03369,0.52664],"element_id":"fa37bc71e569e9fa6e29d76627af96d280c2b13bfb548fdd20af7a1765e919db","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"**四、Top 10用量用户排行**","type":"TableCaption"},{"coordinates":[0.026203,0.534413,0.958289,0.534413,0.959893,0.647287,0.027807,0.647287],"element_id":"27293311abf02020ae64ebec82104dfd5d9b55a71a81cf195a3e3f88a06fe83d","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"ref_element_id":"fa37bc71e569e9fa6e29d76627af96d280c2b13bfb548fdd20af7a1765e919db"},"page_number":1,"sub_type":"borderless","text":"<table border=\\"1\\" ><tr>\\n<td>排名</td>\\n<td>姓名</td>\\n<td>所属部门／二级部门</td>\\n<td>API调用次数</td>\\n<td>Token总量</td>\\n<td>费用（¥）</td>\\n<td>人均费用排名</td>\\n</tr><tr>\\n<td>1<img src=\\"https://web-api.textin.com/ocr_image/external/ba7f3d967ae7e1e0.jpg\\"></td>\\n<td>钱伟豪</td>\\n<td>数据科学部／算法研究组</td>\\n<td>2,450,000</td>\\n<td>485.00亿</td>\\n<td>1,890,000.00</td>\\n<td>1</td>\\n</tr><tr>\\n<td>2<img src=\\"https://web-api.textin.com/ocr_image/external/53bc114224d67844.jpg\\"></td>\\n<td>郑子涵</td>\\n<td>运营部／用户运营组</td>\\n<td>2,280,000</td>\\n<td>420.00亿</td>\\n<td>1,750,000.00</td>\\n<td>2</td>\\n</tr><tr>\\n<td>3<img src=\\"https://web-api.textin.com/ocr_image/external/43d1eafd23a09124.jpg\\"></td>\\n<td>张明远</td>\\n<td>研发部／AI平台组</td>\\n<td>2,150,000</td>\\n<td>395.00亿</td>\\n<td>1,620,000.00</td>\\n<td>3</td>\\n</tr><tr>\\n<td>4<img src=\\"https://web-api.textin.com/ocr_image/external/e18b5369f3912613.jpg\\"></td>\\n<td>陆雨萱</td>\\n<td>运营部／数据运营组</td>\\n<td>1,980,000</td>\\n<td>370.00亿</td>\\n<td>1,480,000.00</td>\\n<td>4</td>\\n</tr><tr>\\n<td>5<img src=\\"https://web-api.textin.com/ocr_image/external/48f179722aeb7feb.jpg\\"></td>\\n<td>周思远</td>\\n<td>市场部／内容营销组</td>\\n<td>1,850,000</td>\\n<td>340.00亿</td>\\n<td>1,420,000.00</td>\\n<td>5</td>\\n</tr><tr>\\n<td>6<img src=\\"https://web-api.textin.com/ocr_image/external/ea746cadc19fece9.jpg\\"></td>\\n<td>赵雨婷</td>\\n<td>产品部／智能产品线</td>\\n<td>1,720,000</td>\\n<td>315.00亿</td>\\n<td>1,280,000.00</td>\\n<td>6</td>\\n</tr><tr>\\n<td>7<img src=\\"https://web-api.textin.com/ocr_image/external/b791b3f37b98a5d8.jpg\\"></td>\\n<td>孙浩然</td>\\n<td>产品部／数据产品线</td>\\n<td>1,580,000</td>\\n<td>290.00亿</td>\\n<td>1,180,000.00</td>\\n<td>7</td>\\n</tr><tr>\\n<td>8<img src=\\"https://web-api.textin.com/ocr_image/external/0c046b802aee1247.jpg\\"></td>\\n<td>李雪琴</td>\\n<td>研发部／基础架构组</td>\\n<td>1,350,000</td>\\n<td>250.00亿</td>\\n<td>1,050,000.00</td>\\n<td>8</td>\\n</tr><tr>\\n<td>9<img src=\\"https://web-api.textin.com/ocr_image/external/0aa4be2f11a1a87a.jpg\\"></td>\\n<td>吴佳琪</td>\\n<td>市场部／品牌设计组</td>\\n<td>980,000</td>\\n<td>185.00亿</td>\\n<td>780,000.00</td>\\n<td>9</td>\\n</tr><tr>\\n<td>10<img src=\\"https://web-api.textin.com/ocr_image/external/ed61a7dd2dacb18a.jpg\\"></td>\\n<td>王建国</td>\\n<td>研发部／测试效能组</td>\\n<td>850,000</td>\\n<td>160.00亿</td>\\n<td>680,000.00</td>\\n<td>11</td>\\n</tr></table>","type":"Table"},{"coordinates":[0.03369,0.648907,0.467914,0.648907,0.467914,0.652146,0.03369,0.652146],"element_id":"6be7c91a4d9f873327966ec3f0f878de7f6002c111cf116bc7ad423f5956cb71","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"注：王建国人均费用排名第11位，因其组内人数较少（18人），人均费用略高于部分排名靠前的用户。","type":"NarrativeText"},{"coordinates":[0.035294,0.673522,0.186096,0.673522,0.186096,0.678057,0.035294,0.678057],"element_id":"6ab79c8789460b2b36aabcec10b91e869d18270409a3e5c1835e95858e4f4ebd","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"**五、项目级用量 Top 10**","type":"TableCaption"},{"coordinates":[0.026738,0.688097,0.963636,0.687126,0.964706,0.795789,0.027273,0.796761],"element_id":"ee9df3e9367c280e8e7f8799dc3d4a2572d455c8b11e8a945cce043f371e4702","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"ref_element_id":"6ab79c8789460b2b36aabcec10b91e869d18270409a3e5c1835e95858e4f4ebd"},"page_number":1,"sub_type":"borderless","text":"<table border=\\"1\\" ><tr>\\n<td>排名</td>\\n<td>项目名称</td>\\n<td>所属部门</td>\\n<td>负责人</td>\\n<td>API调用</td>\\n<td>Token总量</td>\\n<td>费用（¥）</td>\\n<td>主要使用模型</td>\\n</tr><tr>\\n<td><img src=\\"https://web-api.textin.com/ocr_image/external/06956064d4cf02d3.jpg\\"></td>\\n<td>智能客服</td>\\n<td>研发部／AI平台组</td>\\n<td>张明远</td>\\n<td>5,800,000</td>\\n<td>892.00亿</td>\\n<td>3,780,000.00</td>\\n<td>GPT-4/ Claude Opus</td>\\n</tr><tr>\\n<td>2<img src=\\"https://web-api.textin.com/ocr_image/external/f3c6386ca77d6467.jpg\\"></td>\\n<td>大模型微调</td>\\n<td>数据科学部／算法研究组</td>\\n<td>钱伟豪</td>\\n<td>5,200,000</td>\\n<td>785.00亿</td>\\n<td>3,450,000.00</td>\\n<td>Claude Opus/GPT-4</td>\\n</tr><tr>\\n<td>3<img src=\\"https://web-api.textin.com/ocr_image/external/cb6fbea2a2a9ede1.jpg\\"></td>\\n<td>用户画像</td>\\n<td>运营部／用户运营组</td>\\n<td>郑子涵</td>\\n<td>4,800,000</td>\\n<td>720.00亿</td>\\n<td>3,120,000.00</td>\\n<td>GPT-4 / Claude Sonnet</td>\\n</tr><tr>\\n<td>4<img src=\\"https://web-api.textin.com/ocr_image/external/dc31a7ad512f72b6.jpg\\"></td>\\n<td>AI搜索</td>\\n<td>产品部／智能产品线</td>\\n<td>赵雨婷</td>\\n<td>4,200,000</td>\\n<td>620.00亿</td>\\n<td>2,680,000.00</td>\\n<td>GPT-4／通义千问</td>\\n</tr><tr>\\n<td>5<img src=\\"https://web-api.textin.com/ocr_image/external/0f1482ed0eead7af.jpg\\"></td>\\n<td>内容生成</td>\\n<td>市场部／内容营销组</td>\\n<td>周思远</td>\\n<td>4,000,000</td>\\n<td>585.00亿</td>\\n<td>2,540,000.00</td>\\n<td>GPT-4/Claude Sonnet</td>\\n</tr><tr>\\n<td>6</td>\\n<td>精准推送</td>\\n<td>运营部／用户运营组</td>\\n<td>郑子涵</td>\\n<td>3,600,000</td>\\n<td>530.00亿</td>\\n<td>2,300,000.00</td>\\n<td>Claude Opus / GPT-4</td>\\n</tr><tr>\\n<td>7</td>\\n<td>文档理解</td>\\n<td>数据科学部／算法研究组</td>\\n<td>钱伟豪</td>\\n<td>3,400,000</td>\\n<td>500.00亿</td>\\n<td>2,180,000.00</td>\\n<td>GPT-4/Claude Sonnet</td>\\n</tr><tr>\\n<td>8<img src=\\"https://web-api.textin.com/ocr_image/external/f2c2ba459ea83fed.jpg\\"></td>\\n<td>代码助手</td>\\n<td>研发部／AI平台组</td>\\n<td>张明远</td>\\n<td>3,200,000</td>\\n<td>475.00亿</td>\\n<td>2,050,000.00</td>\\n<td>Claude Opus/GPT-4</td>\\n</tr><tr>\\n<td>9<img src=\\"https://web-api.textin.com/ocr_image/external/b777c77519a0fb55.jpg\\"></td>\\n<td>A/B测试平台</td>\\n<td>运营部／数据运营组</td>\\n<td>陆雨萱</td>\\n<td>2,900,000</td>\\n<td>430.00亿</td>\\n<td>1,860,000.00</td>\\n<td>GPT-4／通义千问</td>\\n</tr><tr>\\n<td>10<img src=\\"https://web-api.textin.com/ocr_image/external/bd6fc7e64cf682f4.jpg\\"></td>\\n<td>智能问答</td>\\n<td>产品部／智能产品线</td>\\n<td>赵雨婷</td>\\n<td>2,700,000</td>\\n<td>400.00亿</td>\\n<td>1,730,000.00</td>\\n<td>GPT-4／文心一言4.0</td>\\n</tr></table>","type":"Table"},{"coordinates":[0.03369,0.817328,0.19893,0.817328,0.19893,0.821862,0.03369,0.821862],"element_id":"59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf","metadata":{"category_depth":0,"children_ids":["0d361f4d825cf1d411087be19d366b2a2cd5962a23ca9e85c772196ff835238b","4fc3d5bfc0b7071ebe9802e674ced047934dcbbf4f5ba71163cc5d76e53355f4","0ed120e82f9d23db44283751a04bfe7f8d9ee7ae4d2bb97086eac6a1f2f72028","45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"六、异常告警与优化建议","type":"Title"},{"coordinates":[0.054011,0.835789,0.074332,0.835789,0.074332,0.841457,0.054011,0.841457],"element_id":"0d361f4d825cf1d411087be19d366b2a2cd5962a23ca9e85c772196ff835238b","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":35,"is_continuation":false,"parent_id":"59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf","width":38},"page_number":1,"text":"","type":"Image"},{"coordinates":[0.080214,0.836113,0.249733,0.836113,0.249733,0.839676,0.080214,0.839676],"element_id":"4fc3d5bfc0b7071ebe9802e674ced047934dcbbf4f5ba71163cc5d76e53355f4","metadata":{"category_depth":1,"children_ids":["f85a431352d4bbaee1f2703ebb3ce1facd38e95ddbdae088cc2588bb1196c982","d890fc45661c5a8c08ea3ac8ee007eeb8842d1371b47f005031cb736130c80f4"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf"},"page_number":1,"text":"内容营销组12月API调用激增","type":"Title"},{"coordinates":[0.080214,0.842591,0.793583,0.842591,0.793583,0.84664,0.080214,0.84664],"element_id":"f85a431352d4bbaee1f2703ebb3ce1facd38e95ddbdae088cc2588bb1196c982","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"4fc3d5bfc0b7071ebe9802e674ced047934dcbbf4f5ba71163cc5d76e53355f4"},"page_number":1,"text":"12月份调用720,000次，较11月（680,000次）增长5.9%，较9月（630,000次）增长14.3％。建议确认是否存在非授权批量调用。","type":"NarrativeText"},{"coordinates":[0.054545,0.857814,0.074332,0.857814,0.074332,0.863968,0.054545,0.863968],"element_id":"d890fc45661c5a8c08ea3ac8ee007eeb8842d1371b47f005031cb736130c80f4","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":38,"is_continuation":false,"parent_id":"4fc3d5bfc0b7071ebe9802e674ced047934dcbbf4f5ba71163cc5d76e53355f4","width":37},"page_number":1,"text":"","type":"Image"},{"coordinates":[0.05508,0.858785,0.32139,0.858785,0.32139,0.86251,0.05508,0.86251],"element_id":"0ed120e82f9d23db44283751a04bfe7f8d9ee7ae4d2bb97086eac6a1f2f72028","metadata":{"category_depth":1,"children_ids":["6841538746fc9f3da46a66c24cd07a07c5c0e6558a4075bcd0ce148a98b0f173","40507e4a4de837fac52069b3cd8052b958d0521b70dc11db8a4fe7b2fea5758f"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf"},"page_number":1,"text":"算法研究组 Claude Opus 费用严重超预算","type":"Title"},{"coordinates":[0.079144,0.865101,0.942781,0.865101,0.942781,0.875466,0.079144,0.875466],"element_id":"6841538746fc9f3da46a66c24cd07a07c5c0e6558a4075bcd0ce148a98b0f173","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"0ed120e82f9d23db44283751a04bfe7f8d9ee7ae4d2bb97086eac6a1f2f72028"},"page_number":1,"text":"Claude Opus全年费用612万元，占部门总费用的41.3%，较年初预算超支12％。建议评估是否可将部分任务迁移至Claude Sonnet以降低费用。按干Token 差价0.01元计算，若迁移30％的Opus调用量，预计年节省费用约110万元。","type":"NarrativeText"},{"coordinates":[0.054545,0.886964,0.073262,0.886964,0.073262,0.892794,0.054011,0.892794],"element_id":"40507e4a4de837fac52069b3cd8052b958d0521b70dc11db8a4fe7b2fea5758f","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":36,"is_continuation":false,"parent_id":"0ed120e82f9d23db44283751a04bfe7f8d9ee7ae4d2bb97086eac6a1f2f72028","width":35},"page_number":1,"text":"","type":"Image"},{"coordinates":[0.079144,0.887611,0.301604,0.887611,0.301604,0.891498,0.079144,0.891498],"element_id":"45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c","metadata":{"category_depth":1,"children_ids":["d8dbc7a9e5eb026f2dffa9644d6f105a2dfee671e900b3a7cb4523244e0f7020","41096bd2e917c5105652fad5b6d174762ca9d214f08f430b9857405102eb53bd","cbf50cca607f8489d297a89badd62f1c00bdc880c7728fb13cb80882bb4cb285","0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf"},"page_number":1,"text":"国产模型占比偏低，成本优化空间显著","type":"Title"},{"coordinates":[0.079144,0.894089,0.949198,0.894089,0.949198,0.904453,0.079144,0.904453],"element_id":"d8dbc7a9e5eb026f2dffa9644d6f105a2dfee671e900b3a7cb4523244e0f7020","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c"},"page_number":1,"text":"文心一言4.0千Token均价0.026元，通义千问0.022元，均显著低于GPT-4(0.045元）和Claude Opus(0.044元）。当前国产模型调用次数占比仅18.3%，费用占比仅9.9％。用户运营组文心一言调用占比14.6%，建议全集团层面增加国产模型使用比例以控制成本。","type":"NarrativeText"},{"coordinates":[0.054011,0.915951,0.073797,0.915789,0.073797,0.921619,0.054011,0.921619],"element_id":"41096bd2e917c5105652fad5b6d174762ca9d214f08f430b9857405102eb53bd","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":35,"is_continuation":false,"parent_id":"45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c","width":37},"page_number":1,"text":"","type":"Image"},{"coordinates":[0.054545,0.916275,0.25508,0.916275,0.25508,0.920162,0.054545,0.920162],"element_id":"cbf50cca607f8489d297a89badd62f1c00bdc880c7728fb13cb80882bb4cb285","metadata":{"category_depth":2,"children_ids":["8849f6dbfbe793665ebdb3e6e307d1d8797c264b9ebe06a404316ed8c44890dd","4a6dbe6256f876cf819f1d197fa3470a763b1555bdb625aff77c81a847f29333"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c"},"page_number":1,"text":"数据工程组Q4调用量平稳回落","type":"Title"},{"coordinates":[0.079144,0.923077,0.73262,0.923077,0.73262,0.927126,0.079144,0.927126],"element_id":"8849f6dbfbe793665ebdb3e6e307d1d8797c264b9ebe06a404316ed8c44890dd","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"cbf50cca607f8489d297a89badd62f1c00bdc880c7728fb13cb80882bb4cb285"},"page_number":1,"text":"Q4月度调用量稳定在12-13万次，较Q2巅峰（11.5万次）有所回落，H2整体趋势平稳，属正常波动，无需特殊关注。","type":"NarrativeText"},{"coordinates":[0.05508,0.938947,0.073262,0.938785,0.073797,0.944453,0.05508,0.944453],"element_id":"4a6dbe6256f876cf819f1d197fa3470a763b1555bdb625aff77c81a847f29333","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":34,"is_continuation":false,"parent_id":"cbf50cca607f8489d297a89badd62f1c00bdc880c7728fb13cb80882bb4cb285","width":34},"page_number":1,"text":"","type":"Image"},{"coordinates":[0.080214,0.938947,0.264706,0.938947,0.264706,0.942996,0.080214,0.942996],"element_id":"0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce","metadata":{"category_depth":2,"children_ids":["4a922b435a499e99243cb96ed4fbbac91c8904ac6cf85f7ab6e418f808e737cf","d1a7efcef9dc649c3fcda506755c4144406831abe0d5dd0f423b6d5ba4dc01dc","05ae9dc9fe33dadf39ca2c13203e369a804bd771a0a84782055e86b828cc145e"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c"},"page_number":1,"text":"品牌设计组人均费用全集团最高","type":"Title"},{"coordinates":[0.080214,0.945101,0.930481,0.945101,0.930481,0.955789,0.080214,0.955789],"element_id":"4a922b435a499e99243cb96ed4fbbac91c8904ac6cf85f7ab6e418f808e737cf","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce"},"page_number":1,"text":"品牌设计组15人，人均费用68,200元／年，为全公司最高（全公司均值29,000元／年，为2.35倍）。建议审查是否存在个人账号混用或多项目共用同-API Key的情况。","type":"NarrativeText"},{"coordinates":[0.334759,0.981377,0.664171,0.981377,0.664171,0.984453,0.334759,0.984453],"element_id":"d1a7efcef9dc649c3fcda506755c4144406831abe0d5dd0f423b6d5ba4dc01dc","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce"},"page_number":1,"text":"制表日期：2026-01-10｜报表版本：v2.3｜下次报表生成日期：2026-02-05","type":"NarrativeText"},{"coordinates":[0.360428,0.986073,0.637433,0.986073,0.637433,0.989312,0.360428,0.989312],"element_id":"05ae9dc9fe33dadf39ca2c13203e369a804bd771a0a84782055e86b828cc145e","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce"},"page_number":1,"text":"原始数据来源于集团AI网关日志系统｜本报表仅供内部管理使用","type":"NarrativeText"}],"file_id":"49a0ab97a3c14bf18b88b8d49cbae812","history_info":{"cloud_ocr":0,"ctime":"1781096956","id":5493532,"img_name":"某某公司AI用量报表.png","img_uri":"https://web-api.textin.com/image/download?filename=9325751c5ddd4efba30783a88b.png","ocr_status":1,"status":0,"thumbnail":"","utime":"1781096956"},"job_id":"9891a8e8b7cb4d1f88799b76527e0fc5","markdown":"<!– **一、各部门Token用量与费用明细** 核心报表 –>\\n\\n<table border=\\"1\\" ><tr>\\n<td rowspan=\\"2\\">部门</td>\\n<td rowspan=\\"2\\">二级部门</td>\\n<td rowspan=\\"2\\">负责人</td>\\n<td rowspan=\\"2\\">人数</td>\\n<td rowspan=\\"2\\">主要模型</td>\\n<td colspan=\\"2\\">Token 用量</td>\\n<td rowspan=\\"2\\">费用（¥）</td>\\n<td rowspan=\\"2\\">预算（¥）</td>\\n<td rowspan=\\"2\\">执行率</td>\\n</tr><tr>\\n<td>输入Token</td>\\n<td>输出Token</td>\\n</tr><tr>\\n<td rowspan=\\"4\\">研发部</td>\\n<td>合计·95 人</td>\\n<td></td>\\n<td>95</td>\\n<td>-</td>\\n<td>2,033.40 亿</td>\\n<td>251.00 亿</td>\\n<td>6,535,789.50</td>\\n<td>7,200,000.00</td>\\n<td>90.8%</td>\\n</tr><tr>\\n<td>AI 平台组</td>\\n<td>张明远</td>\\n<td>45</td>\\n<td>GPT-4/Claude Opus／文心一言4.0</td>\\n<td>1,796.00 亿</td>\\n<td>251.00 亿</td>\\n<td>6,205,789.50</td>\\n<td>6,800,000.00</td>\\n<td>91.3%</td>\\n</tr><tr>\\n<td>基础架构<br>组</td>\\n<td>李雪琴</td>\\n<td>32</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>770.00 亿</td>\\n<td>115.00 亿</td>\\n<td>2,670,000.00</td>\\n<td>3,000,000.00</td>\\n<td>89.0%</td>\\n</tr><tr>\\n<td>测试效能<br>组</td>\\n<td>王建国</td>\\n<td>18</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>198.00 亿</td>\\n<td>37.00 亿</td>\\n<td>690,000.00</td>\\n<td>800,000.00</td>\\n<td>86.3%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">产品部</td>\\n<td>合计·50 人</td>\\n<td></td>\\n<td>50</td>\\n<td>-</td>\\n<td>1,780.00 亿</td>\\n<td>240.00 亿</td>\\n<td>5,900,000.00</td>\\n<td>6,500,000.00</td>\\n<td>90.8%</td>\\n</tr><tr>\\n<td>智能产品<br>线</td>\\n<td>赵雨婷</td>\\n<td>28</td>\\n<td>GPT-4/Claude Opus／通义千问</td>\\n<td>1,150.00 亿</td>\\n<td>195.00 亿</td>\\n<td>4,240,000.00</td>\\n<td>4,600,000.00</td>\\n<td>92.2%</td>\\n</tr><tr>\\n<td>数据产品<br>线</td>\\n<td>孙浩然</td>\\n<td>22</td>\\n<td>GPT-4 / Claude Sonnet／文心一言4.0</td>\\n<td>740.00 亿</td>\\n<td>115.00 亿</td>\\n<td>2,330,000.00</td>\\n<td>2,600,000.00</td>\\n<td>89.6%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">市场部</td>\\n<td>合计·50 人</td>\\n<td></td>\\n<td>50</td>\\n<td>-</td>\\n<td>1,195.00 亿</td>\\n<td>206.00 亿</td>\\n<td>5,453,000.00</td>\\n<td>5,800,000.00</td>\\n<td>94.0%</td>\\n</tr><tr>\\n<td>内容营销<br>组</td>\\n<td>周思远</td>\\n<td>35</td>\\n<td>GPT-4/ Claude Opus/Claude Sonnet</td>\\n<td>950.00 亿</td>\\n<td>257.00 亿</td>\\n<td>5,450,000.00</td>\\n<td>5,800,000.00</td>\\n<td>94.0%</td>\\n</tr><tr>\\n<td>品牌设计<br>组</td>\\n<td>吴佳琪</td>\\n<td>15</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>185.00 亿</td>\\n<td>84.00 亿</td>\\n<td>1,023,000.00</td>\\n<td>1,200,000.00</td>\\n<td>85.3%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">运营部</td>\\n<td>合计·65 人</td>\\n<td></td>\\n<td>65</td>\\n<td>-</td>\\n<td>1,990.00 亿</td>\\n<td>485.00 亿</td>\\n<td>10,800,000.00</td>\\n<td>12,000,000.00</td>\\n<td>90.0%</td>\\n</tr><tr>\\n<td>用户运营<br>组</td>\\n<td>郑子涵</td>\\n<td>40</td>\\n<td>GPT-4/Claude Opus/Claude Sonnet／文心一言4.0</td>\\n<td>1,990.00 亿</td>\\n<td>785.00 亿</td>\\n<td>10,550,000.00</td>\\n<td>11,500,000.00</td>\\n<td>91.7%</td>\\n</tr><tr>\\n<td>数据运营<br>组</td>\\n<td>陆雨萱</td>\\n<td>25</td>\\n<td>GPT-4/Claude Opus／通义千问</td>\\n<td>880.00 亿</td>\\n<td>425.00 亿</td>\\n<td>5,010,000.00</td>\\n<td>5,500,000.00</td>\\n<td>91.1%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">数据科学<br>部</td>\\n<td>合计·58 人</td>\\n<td></td>\\n<td>58</td>\\n<td>-</td>\\n<td>2,940.00 亿</td>\\n<td>372.00 亿</td>\\n<td>14,814,000.00</td>\\n<td>16,000,000.00</td>\\n<td>92.6%</td>\\n</tr><tr>\\n<td>算法研究<br>组</td>\\n<td>钱伟豪</td>\\n<td>38</td>\\n<td>GPT-4 / Claude Opus / Claude Sonnet／文心一言4.0／<br>通义千问</td>\\n<td>2,950.00 亿</td>\\n<td>467.00 亿</td>\\n<td>14,810,000.00</td>\\n<td>16,000,000.00</td>\\n<td>92.6%</td>\\n</tr><tr>\\n<td>数据工程<br>组</td>\\n<td>陈俊杰</td>\\n<td>20</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>248.00 亿</td>\\n<td>70.00 亿</td>\\n<td>1,134,000.00</td>\\n<td>1,300,000.00</td>\\n<td>87.2%</td>\\n</tr><tr>\\n<td rowspan=\\"2\\">财务部</td>\\n<td>合计·12 人</td>\\n<td></td>\\n<td>12</td>\\n<td>-</td>\\n<td>90.00 亿</td>\\n<td>27.50 亿</td>\\n<td>434,000.00</td>\\n<td>500,000.00</td>\\n<td>86.8%</td>\\n</tr><tr>\\n<td>财务分析<br>组</td>\\n<td>林妙音</td>\\n<td>12</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>90.00 亿</td>\\n<td>27.50 亿</td>\\n<td>434,000.00</td>\\n<td>500,000.00</td>\\n<td>86.8%</td>\\n</tr><tr>\\n<td>集团合计</td>\\n<td></td>\\n<td></td>\\n<td></td>\\n<td>-</td>\\n<td>10,028.40<br>亿</td>\\n<td>1,906.50<br>亿</td>\\n<td>¥43,936,789.50</td>\\n<td>¥48,000,000.00</td>\\n<td>91.5%</td>\\n</tr></table>\\n\\n**二、各模型用量与费用对比**\\n\\n<table border=\\"1\\" ><tr>\\n<td>模型</td>\\n<td>调用次数</td>\\n<td>Token 小计</td>\\n<td>总费用（¥）</td>\\n<td>干Token 均价（¥）</td>\\n<td>费用占比</td>\\n</tr><tr>\\n<td>GPT-4</td>\\n<td>15,680,000</td>\\n<td>4,720.00 亿</td>\\n<td>21,278,000.00</td>\\n<td>0.045</td>\\n<td>43.2%</td>\\n</tr><tr>\\n<td>Claude Opus</td>\\n<td>8,950,000</td>\\n<td>3,515.50 亿</td>\\n<td>15,645,000.00</td>\\n<td>0.044</td>\\n<td>31.8%</td>\\n</tr><tr>\\n<td>Claude Sonnet</td>\\n<td>8,720,000</td>\\n<td>2,034.00 亿</td>\\n<td>6,885,000.00</td>\\n<td>0.034</td>\\n<td>14.0%</td>\\n</tr><tr>\\n<td>文心一言4.0</td>\\n<td>4,670,000</td>\\n<td>1,267.20 亿</td>\\n<td>3,310,000.00</td>\\n<td>0.026</td>\\n<td>6.7%</td>\\n</tr><tr>\\n<td>通义千<br>问</td>\\n<td>2,960,000</td>\\n<td>724.50 亿</td>\\n<td>1,570,000.00</td>\\n<td>0.022</td>\\n<td>3.2%</td>\\n</tr><tr>\\n<td>合计</td>\\n<td>40,980,000</td>\\n<td>12,261.20<br>亿</td>\\n<td>48,688,000.00</td>\\n<td>-</td>\\n<td>100%</td>\\n</tr></table>\\n\\n**三、月度趋势（2025年）**\\n\\n<table border=\\"1\\" ><tr>\\n<td>月份</td>\\n<td>调用次数</td>\\n<td>Token 总量</td>\\n<td>费用（¥）</td>\\n<td>环比（费用）</td>\\n</tr><tr>\\n<td>1 月</td>\\n<td>2,575,000</td>\\n<td>485.00 亿</td>\\n<td>2,722,000.00</td>\\n<td>-</td>\\n</tr><tr>\\n<td>2 月</td>\\n<td>2,472,000</td>\\n<td>464.00 亿</td>\\n<td>2,605,000.00</td>\\n<td>-4.3%<br><img src=\\"https://web-api.textin.com/ocr_image/external/2a10b2c1397dab4b.jpg\\"></td>\\n</tr><tr>\\n<td>3 月</td>\\n<td>3,038,000</td>\\n<td>572.00 亿</td>\\n<td>3,205,000.00</td>\\n<td>+23.0%</td>\\n</tr><tr>\\n<td>4 月</td>\\n<td>3,263,000</td>\\n<td>614.00 亿</td>\\n<td>3,446,000.00</td>\\n<td>+7.5%</td>\\n</tr><tr>\\n<td>5 月</td>\\n<td>3,464,000</td>\\n<td>652.00 亿</td>\\n<td>3,659,000.00</td>\\n<td>+6.2%</td>\\n</tr><tr>\\n<td>6 月</td>\\n<td>3,630,000</td>\\n<td>683.00 亿</td>\\n<td>3,839,000.00</td>\\n<td>+4.9%</td>\\n</tr><tr>\\n<td>7 月</td>\\n<td>3,538,000</td>\\n<td>665.00 亿</td>\\n<td>3,733,000.00</td>\\n<td>-2.8%</td>\\n</tr><tr>\\n<td>8 月</td>\\n<td>3,721,000</td>\\n<td>700.00 亿</td>\\n<td>3,936,000.00</td>\\n<td>+5.4%</td>\\n</tr><tr>\\n<td>9 月</td>\\n<td>3,646,000</td>\\n<td>685.00 亿</td>\\n<td>3,850,000.00</td>\\n<td>-2.2%</td>\\n</tr><tr>\\n<td>10 月</td>\\n<td>3,839,000</td>\\n<td>721.00 亿</td>\\n<td>4,054,000.00</td>\\n<td>+5.3%</td>\\n</tr><tr>\\n<td>11 月</td>\\n<td>3,928,000</td>\\n<td>739.00 亿</td>\\n<td>4,153,000.00</td>\\n<td>+2.4%</td>\\n</tr><tr>\\n<td>12 月</td>\\n<td>4,146,000</td>\\n<td>780.00 亿</td>\\n<td>4,388,000.00</td>\\n<td>+5.7%</td>\\n</tr><tr>\\n<td>累计</td>\\n<td>42,260,000</td>\\n<td>7,760.00 亿</td>\\n<td>43,590,000.00</td>\\n<td>-</td>\\n</tr></table>\\n\\n<!– 440 385 月度费用 366 384 373 394 345 321 400 272 261 (A) 350 旺邮 300 250 1月 2月 3月 4月 5月 6月 7月 8月 9月 10月 11月 12月 –>\\n\\n\\n**四、Top 10用量用户排行**\\n\\n<table border=\\"1\\" ><tr>\\n<td>排名</td>\\n<td>姓名</td>\\n<td>所属部门／二级部门</td>\\n<td>API 调用次数</td>\\n<td>Token 总量</td>\\n<td>费用（¥）</td>\\n<td>人均费用排名</td>\\n</tr><tr>\\n<td>1<br><img src=\\"https://web-api.textin.com/ocr_image/external/ba7f3d967ae7e1e0.jpg\\"></td>\\n<td>钱伟豪</td>\\n<td>数据科学部／算法研究组</td>\\n<td>2,450,000</td>\\n<td>485.00 亿</td>\\n<td>1,890,000.00</td>\\n<td>1</td>\\n</tr><tr>\\n<td>2<br><img src=\\"https://web-api.textin.com/ocr_image/external/53bc114224d67844.jpg\\"></td>\\n<td>郑子涵</td>\\n<td>运营部／用户运营组</td>\\n<td>2,280,000</td>\\n<td>420.00 亿</td>\\n<td>1,750,000.00</td>\\n<td>2</td>\\n</tr><tr>\\n<td>3<br><img src=\\"https://web-api.textin.com/ocr_image/external/43d1eafd23a09124.jpg\\"></td>\\n<td>张明远</td>\\n<td>研发部／AI 平台组</td>\\n<td>2,150,000</td>\\n<td>395.00 亿</td>\\n<td>1,620,000.00</td>\\n<td>3</td>\\n</tr><tr>\\n<td>4<br><img src=\\"https://web-api.textin.com/ocr_image/external/e18b5369f3912613.jpg\\"></td>\\n<td>陆雨萱</td>\\n<td>运营部／数据运营组</td>\\n<td>1,980,000</td>\\n<td>370.00 亿</td>\\n<td>1,480,000.00</td>\\n<td>4</td>\\n</tr><tr>\\n<td>5<br><img src=\\"https://web-api.textin.com/ocr_image/external/48f179722aeb7feb.jpg\\"></td>\\n<td>周思远</td>\\n<td>市场部／内容营销组</td>\\n<td>1,850,000</td>\\n<td>340.00 亿</td>\\n<td>1,420,000.00</td>\\n<td>5</td>\\n</tr><tr>\\n<td>6<br><img src=\\"https://web-api.textin.com/ocr_image/external/ea746cadc19fece9.jpg\\"></td>\\n<td>赵雨婷</td>\\n<td>产品部／智能产品线</td>\\n<td>1,720,000</td>\\n<td>315.00 亿</td>\\n<td>1,280,000.00</td>\\n<td>6</td>\\n</tr><tr>\\n<td>7<br><img src=\\"https://web-api.textin.com/ocr_image/external/b791b3f37b98a5d8.jpg\\"></td>\\n<td>孙浩然</td>\\n<td>产品部／数据产品线</td>\\n<td>1,580,000</td>\\n<td>290.00 亿</td>\\n<td>1,180,000.00</td>\\n<td>7</td>\\n</tr><tr>\\n<td>8<br><img src=\\"https://web-api.textin.com/ocr_image/external/0c046b802aee1247.jpg\\"></td>\\n<td>李雪琴</td>\\n<td>研发部／基础架构组</td>\\n<td>1,350,000</td>\\n<td>250.00 亿</td>\\n<td>1,050,000.00</td>\\n<td>8</td>\\n</tr><tr>\\n<td>9<br><img src=\\"https://web-api.textin.com/ocr_image/external/0aa4be2f11a1a87a.jpg\\"></td>\\n<td>吴佳琪</td>\\n<td>市场部／品牌设计组</td>\\n<td>980,000</td>\\n<td>185.00 亿</td>\\n<td>780,000.00</td>\\n<td>9</td>\\n</tr><tr>\\n<td>10<br><img src=\\"https://web-api.textin.com/ocr_image/external/ed61a7dd2dacb18a.jpg\\"></td>\\n<td>王建国</td>\\n<td>研发部／测试效能组</td>\\n<td>850,000</td>\\n<td>160.00 亿</td>\\n<td>680,000.00</td>\\n<td>11</td>\\n</tr></table>\\n\\n注：王建国人均费用排名第11位，因其组内人数较少（18人），人均费用略高于部分排名靠前的用户。\\n\\n**五、项目级用量 Top 10**\\n\\n<table border=\\"1\\" ><tr>\\n<td>排名</td>\\n<td>项目名称</td>\\n<td>所属部门</td>\\n<td>负责人</td>\\n<td>API 调用</td>\\n<td>Token 总量</td>\\n<td>费用（¥）</td>\\n<td>主要使用模型</td>\\n</tr><tr>\\n<td><img src=\\"https://web-api.textin.com/ocr_image/external/06956064d4cf02d3.jpg\\"></td>\\n<td>智能客服</td>\\n<td>研发部／AI 平台组</td>\\n<td>张明远</td>\\n<td>5,800,000</td>\\n<td>892.00 亿</td>\\n<td>3,780,000.00</td>\\n<td>GPT-4/ Claude Opus</td>\\n</tr><tr>\\n<td>2<br><img src=\\"https://web-api.textin.com/ocr_image/external/f3c6386ca77d6467.jpg\\"></td>\\n<td>大模型微调</td>\\n<td>数据科学部／算法研究组</td>\\n<td>钱伟豪</td>\\n<td>5,200,000</td>\\n<td>785.00 亿</td>\\n<td>3,450,000.00</td>\\n<td>Claude Opus/GPT-4</td>\\n</tr><tr>\\n<td>3<br><img src=\\"https://web-api.textin.com/ocr_image/external/cb6fbea2a2a9ede1.jpg\\"></td>\\n<td>用户画像</td>\\n<td>运营部／用户运营组</td>\\n<td>郑子涵</td>\\n<td>4,800,000</td>\\n<td>720.00 亿</td>\\n<td>3,120,000.00</td>\\n<td>GPT-4 / Claude Sonnet</td>\\n</tr><tr>\\n<td>4<br><img src=\\"https://web-api.textin.com/ocr_image/external/dc31a7ad512f72b6.jpg\\"></td>\\n<td>AI 搜索</td>\\n<td>产品部／智能产品线</td>\\n<td>赵雨婷</td>\\n<td>4,200,000</td>\\n<td>620.00 亿</td>\\n<td>2,680,000.00</td>\\n<td>GPT -4 ／通义千问</td>\\n</tr><tr>\\n<td>5<br><img src=\\"https://web-api.textin.com/ocr_image/external/0f1482ed0eead7af.jpg\\"></td>\\n<td>内容生成</td>\\n<td>市场部／内容营销组</td>\\n<td>周思远</td>\\n<td>4,000,000</td>\\n<td>585.00 亿</td>\\n<td>2,540,000.00</td>\\n<td>GPT-4/Claude Sonnet</td>\\n</tr><tr>\\n<td>6</td>\\n<td>精准推送</td>\\n<td>运营部／用户运营组</td>\\n<td>郑子涵</td>\\n<td>3,600,000</td>\\n<td>530.00 亿</td>\\n<td>2,300,000.00</td>\\n<td>Claude Opus / GPT-4</td>\\n</tr><tr>\\n<td>7</td>\\n<td>文档理解</td>\\n<td>数据科学部／算法研究组</td>\\n<td>钱伟豪</td>\\n<td>3,400,000</td>\\n<td>500.00 亿</td>\\n<td>2,180,000.00</td>\\n<td>GPT-4/Claude Sonnet</td>\\n</tr><tr>\\n<td>8<br><img src=\\"https://web-api.textin.com/ocr_image/external/f2c2ba459ea83fed.jpg\\"></td>\\n<td>代码助手</td>\\n<td>研发部／AI 平台组</td>\\n<td>张明远</td>\\n<td>3,200,000</td>\\n<td>475.00 亿</td>\\n<td>2,050,000.00</td>\\n<td>Claude Opus/GPT-4</td>\\n</tr><tr>\\n<td>9<br><img src=\\"https://web-api.textin.com/ocr_image/external/b777c77519a0fb55.jpg\\"></td>\\n<td>A /B 测试平台</td>\\n<td>运营部／数据运营组</td>\\n<td>陆雨萱</td>\\n<td>2,900,000</td>\\n<td>430.00 亿</td>\\n<td>1,860,000.00</td>\\n<td>GPT -4 ／通义千问</td>\\n</tr><tr>\\n<td>10<br><img src=\\"https://web-api.textin.com/ocr_image/external/bd6fc7e64cf682f4.jpg\\"></td>\\n<td>智能问答</td>\\n<td>产品部／智能产品线</td>\\n<td>赵雨婷</td>\\n<td>2,700,000</td>\\n<td>400.00 亿</td>\\n<td>1,730,000.00</td>\\n<td>GPT -4 ／文心一言4 . 0</td>\\n</tr></table>\\n\\n# 六、异常告警与优化建议\\n\\n\\n\\n\\n## 内容营销组12月API调用激增\\n\\n12月份调用720,000次，较11月（680,000次）增长5.9%，较9月（630,000次）增长14.3％。建议确认是否存在非授权批量调用。\\n\\n\\n\\n\\n## 算法研究组 Claude Opus 费用严重超预算\\n\\nClaude Opus全年费用612万元，占部门总费用的41.3%，较年初预算超支12％。建议评估是否可将部分任务迁移至Claude Sonnet以降低费用。按干Token 差价0.01元计算，若迁移30％的Opus调用量，预计年节省费用约110万元。\\n\\n\\n\\n\\n## 国产模型占比偏低，成本优化空间显著\\n\\n文心一言4.0千Token均价0.026元，通义千问0.022元，均显著低于GPT-4(0.045元）和Claude Opus(0.044元）。当前国产模型调用次数占比仅18.3%，费用占比仅9.9％。用户运营组文心一言调用占比14.6%，建议全集团层面增加国产模型使用比例以控制成本。\\n\\n<!– ☒ –>\\n\\n\\n### 数据工程组Q4调用量平稳回落\\n\\nQ4月度调用量稳定在12-13万次，较Q2巅峰（11.5万次）有所回落，H2整体趋势平稳，属正常波动，无需特殊关注。\\n\\n\\n\\n\\n### 品牌设计组人均费用全集团最高\\n\\n品牌设计组15人，人均费用68,200元／年，为全公司最高（全公司均值29,000元／年，为2.35倍）。建议审查是否存在个人账号混用或多项目共用同-API Key的情况。\\n\\n制表日期：2026-01-10｜报表版本：v2.3｜下次报表生成日期：2026-02-05\\n\\n原始数据来源于集团AI网关日志系统｜本报表仅供内部管理使用\\n\\n","metadata":{"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"filename":"某某公司AI用量报表.png","filetype":"image/png","page_count":1},"pages":[{"angle":0,"element_ids":["b0b28b7598462b1dc35628df9bd3086dbdfa0c4f15fea70bb9a4e328f07e137a","005d44743d4991491f5d4bada074ab5f22e4466730fc16d1bf18d30f8551f7d3","f7fe5462db30f641fafe7ab4f6c144c9310937325cb997d94fc507ca257ec08e","913ee7e62fc92de95e4d7c80b9a84ded6ef77ceb22c86649a7f556f8cad5dd45","74c3526f3d69b8129067aaf8df191f4fdba8d7e9bab184723712ce93b7ed92a1","151a7f0d9eecc434cbe14a1b79375060f837ec40226c689e12e885d3692ebeba","3718ef78786ac3fe0afc9a7ea16d69c0688a0739cfa32f5448e8f0dfa32e295b","fa37bc71e569e9fa6e29d76627af96d280c2b13bfb548fdd20af7a1765e919db","27293311abf02020ae64ebec82104dfd5d9b55a71a81cf195a3e3f88a06fe83d","6be7c91a4d9f873327966ec3f0f878de7f6002c111cf116bc7ad423f5956cb71","6ab79c8789460b2b36aabcec10b91e869d18270409a3e5c1835e95858e4f4ebd","ee9df3e9367c280e8e7f8799dc3d4a2572d455c8b11e8a945cce043f371e4702","59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf","0d361f4d825cf1d411087be19d366b2a2cd5962a23ca9e85c772196ff835238b","4fc3d5bfc0b7071ebe9802e674ced047934dcbbf4f5ba71163cc5d76e53355f4","f85a431352d4bbaee1f2703ebb3ce1facd38e95ddbdae088cc2588bb1196c982","d890fc45661c5a8c08ea3ac8ee007eeb8842d1371b47f005031cb736130c80f4","0ed120e82f9d23db44283751a04bfe7f8d9ee7ae4d2bb97086eac6a1f2f72028","6841538746fc9f3da46a66c24cd07a07c5c0e6558a4075bcd0ce148a98b0f173","40507e4a4de837fac52069b3cd8052b958d0521b70dc11db8a4fe7b2fea5758f","45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c","d8dbc7a9e5eb026f2dffa9644d6f105a2dfee671e900b3a7cb4523244e0f7020","41096bd2e917c5105652fad5b6d174762ca9d214f08f430b9857405102eb53bd","cbf50cca607f8489d297a89badd62f1c00bdc880c7728fb13cb80882bb4cb285","8849f6dbfbe793665ebdb3e6e307d1d8797c264b9ebe06a404316ed8c44890dd","4a6dbe6256f876cf819f1d197fa3470a763b1555bdb625aff77c81a847f29333","0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce","4a922b435a499e99243cb96ed4fbbac91c8904ac6cf85f7ab6e418f808e737cf","d1a7efcef9dc649c3fcda506755c4144406831abe0d5dd0f423b6d5ba4dc01dc","05ae9dc9fe33dadf39ca2c13203e369a804bd771a0a84782055e86b828cc145e"],"page_height":6175,"page_image_url":"https://web-api.textin.com/ocr_image/external/965ea182ea1adbff.jpg","page_number":1,"page_width":1870,"status":"Success"}],"schema_version":"1.3.1","success_count":1,"summary":{"duration_ms":3119}}

关键下游场景：

• RAG 问答： header_tree 让 LLM 能精确回答"项目级用量 Top 10"，而不是模糊地匹配"费用"列
• 知识库构建： footnotes 和 unit 字段保留了数据的完整上下文，避免"拿到正确数字做出错误解读"
• 数据入库： 标准 JSON Schema 可直接对接数据库写入流程，无需二次清洗

准确回答，无一字错漏 👍

六、总结

回到最初的问题：表格解析结果到手了，怎么判断能不能用？

一、引言：复杂表格解析的隐形难题

在处理大批量文档中的表格数据时，我们团队长期被低效的解析质量验证方式和参差不齐的解析工具所困扰——接入过通用OCR API，也测试过多款开源解析方案，却始终没能解决复杂表格（多层表头、合并单元格、跨页续表、嵌套子表）的解析准确性问题。随机抽检的“肉眼筛查”模式，既无法覆盖所有风险，又极易让结构错位、字段归属错误的表格数据流入RAG系统、知识库或风控模型，导致下游业务决策失误。

为了评估当前主流表格解析方案在真实生产场景中的表现，我们选取了市面上关注度较高的几款工具进行横向测试，包括合合信息推出的 TextIn xParse、PaddleOCR、MinerU 等。

**本次测评将从结构、关系、内容准等多个维度展开，并结合实际复杂表格案例，观察不同方案在解析效果和可用性上的差异。**究竟哪种方案更适合落地到企业级应用场景？接下来我们直接看结果。

二、三维度自查清单：系统性地判断解析结果是否可用

这套检查方法将表格解析质量分为三个层次，检查成本依次递增：

自查策略： 先看结构（一眼可见），再看关系（需要对照原文），最后抽查内容（高风险区域抽样）。

2.1 维度一：结构对不对

结构是表格解析的第一关。如果结构层面出了问题，后续的关系归属和内容精度就没有讨论的意义。

结构自查围绕四个问题：

🔬** 实测案例1：复杂合并单元格工程表**

业务场景： 某基坑支护设计规范表，表格同时存在大面积行合并、列合并和跨层级表头。单个单元格内容往往需要关联多行、多列信息才能完整理解其含义。对于表格解析系统而言，不仅要恢复表格结构，更要准确保留“结构类型—安全等级—适用条件”之间的层级关系，否则容易出现字段挂错、内容归属错误等问题。

原表特征：

• Paddle OCR:

测试结果：表格区域错误，合并单元格识别有误

出现行列合并关系丢失，原图是高度结构化的多层嵌套表格（有大合并单元格、不同层级的子表格），识别后变成了松散的文本块；其次表头混乱，原图最上方的“结构类型”“安全等级”“适用条件”等表头基本丢失或错位。

• MinerU:

测试结果：整体表格结构识别正确，多级表头、合并单元格及字段归属关系均得到较好保留，能够还原原始表格的层级结构。

但在细节识别上仍存在一定问题，右侧序号1-5每行的文字推挤成一块，可读性较弱。

• TextIn xParse

测试结果：整体表格结构识别正确，多级表头、合并单元格及字段归属关系均得到较好保留，能够还原原始表格的层级结构。

细节部分出现数字识别错误，对比MinerU对右侧大段内容内容的识别情况，每行显示清晰。

2.2 维度二：关系对不对

关系层比结构层更隐蔽，但直接影响数据的业务含义。你拿到的数据可能行列数量都对、合并单元格也在，但某个数字挂到了错误的表头上——这比结构错误更难发现，危害也更大。

关系自查的核心问题：

🔬** 实测案例：嵌套多个表的劳务报酬审批单**

业务场景： 某企业员工竞业补贴审批单，主表中嵌套了两张独立的追踪表，分别用于记录2025年和2026年的补贴回收情况。两张追踪表均包含独立标题、月份维度及金额数据，需要准确识别表格边界、层级关系以及金额对应月份，避免出现表格合并或数据错位的问题。

原表特征：

实测对比：

• MinerU：

测试结果：嵌套关系丢失

原文中包含两个独立的嵌套表：“2025年收回竞业补贴追踪表”和“2026年收回竞业补贴追踪表”，两张表均有各自的标题、表头和数据区域。

但在解析结果中，系统未能识别两张表之间的边界，将两个嵌套表直接合并成了一张连续表格。其中，第一张表的标题完全丢失，第二张表的标题则被当作普通内容插入到了表格中间。

• PaddleOCR：

测试结果：嵌套关系丢失

同上，原文中包含两个独立的嵌套表，但解析结果未能识别表格边界，将两张表直接拼接在一起，导致嵌套结构丢失；原始标题也未能识别。

此外，表格中的数据归属出现错误。原本应位于11月份列下的金额“200”，被错误识别到了10月份列，导致数据位置发生偏移。

这类错误看似只是列位置偏移，但实际会直接影响数据准确性。尤其是在财务、审计、经营分析等场景中，金额与时间维度的对应关系至关重要，一旦月份与金额错位，可能导致统计结果失真、趋势判断错误，甚至影响后续业务决策。

• TextIn xParse：

测试结果：识别正确

使用TextIn Xparse的vlm解析模式，两个嵌套表均被正确识别和拆分，标题、表头及数据位置保持一致，金额与月份对应关系准确，较好地还原了原始表格结构。

2.3 维度三：内容对不对

如果结构和关系都已通过检查，内容层通常属于低概率出错区。但某些高风险区域仍需重点抽查。

内容自查重点关注：

🔬** 实测案例：审计底稿中的密集数字表格**

业务场景： 随便绘制的AI使用量统计表，包含大量金额数字、百分比、以及脚注小字。表格为扫描件，底部备注使用 7pt 小号字体，且页面有浅色水印覆盖。

原表特征：

文字密集，有些文字模糊不清，有盖章，有签字，有单元格合并，批注等。

注：下图数据仅是随机填写的测试数据，纯属虚构，如有雷同，联系博主

对比差异：

xParse表现稳定，paddle_ocr和mineru少量数字千分号丢失错误，少量运算符号识别丢失错误。

结论： 内容层的错误往往是"细节杀手"——一个丢失的负号意味着坏账变成了收益，一个丢失的百分号意味着比例变成了整数。xParse 在密集数字和低对比度小字两个高风险区域的表现稳定，尤其是对扫描件中小号字体的处理明显优于对比方案。

三、TextIn xParse 产品初体验

3.1 产品定位

TextIn xParse 是合合信息旗下面向 LLM 与 Agent 的文档智能基础设施。它的核心能力是将复杂文档（PDF、扫描件、图片）转化为干净、一致、可用于 AI 的结构化数据，适用于知识库构建、RAG 系统、或其他 AI 工作流的下游应用。

与通用 OCR 工具不同，xParse 专门针对复杂表格做了深度优化——不只是"识别文字"，而是理解表格的结构和语义。

3.2 产品体验流程

访问 https://www.textin.com/register/code/VT4KAZ，登录后，可以直接体验表格解析的全流程。整体操作路径非常简洁，操作流程共三步：

第一步：上传文档。 支持 PDF、PNG、JPG 等格式，可直接拖拽或点击上传。我们选择一张包含复杂表格的 PDF 页面进行测试。

截图描述：展示文件上传区域，支持拖拽上传或本地文件选择。上传区域下方显示支持的文件格式列表（PDF/PNG/JPG），以及单文件大小限制说明。已上传的文件以缩略图形式显示在左侧列表中。

第二步：自动解析。 xParse 提供多种解析模式，针对表格场景可选择"表格解析"模式。

第三步：查看解析结果。 点击"开始解析"后，右侧面板实时渲染解析结果——包括 JSON 结构化数据和可视化表格渲染等多种解析结果。这个"原文 ↔ 解析结果"并排展示的方式，让比对变得直观。

第四步：可以导出常用的格式。解析后一键导出MD、txt、excel、word，格式转换没烦恼！

第五步：加入知识库。准确快速回答你想知道的相关数据信息！

体验小结： 从上传到拿到结构化结果，整个过程不超过 30 秒。界面的"原文 ↔ 解析结果"同屏对比设计在实际使用中非常实用，不需要在多个窗口之间来回切换。

四、python调用api解析文档，三步搞定

第一步：获取API KEY

使用文档解析API处理文档时，您需要先获取API Key。请先登录后前往 TextIn工作台 – 账号与开发者信息 获取您的 x-ti-app-id 和 x-ti-secret-code 。点击获取：https://www.textin.com/console/dashboard/setting

第二步：代码里面改上自己的 x-ti-app-id 和 x-ti-secret-code

import json
import requests
from typing import Dict, Optional
from dataclasses import dataclass

class OCRClient:
def __init__(self, app_id: str, secret_code: str):
self.app_id = app_id
self.secret_code = secret_code

def recognize(self, file_content: bytes, options: dict) -> str:
# 构建请求参数
params = {}
for key, value in options.items():
params[key] = str(value)

# 设置请求头
headers = {
"x-ti-app-id": self.app_id,
"x-ti-secret-code": self.secret_code,
"x-ti-client-source": "sample-code-v1.0",
# 方式一：读取本地文件
"Content-Type": "application/octet-stream"
# 方式二：使用URL方式
# "Content-Type": "text/plain"
}

# 发送请求
response = requests.post(
f"https://api.textin.com/ai/service/v1/pdf_to_markdown",
params=params,
headers=headers,
data=file_content
)

# 检查响应状态
response.raise_for_status()
return response.text

def main():
# 创建客户端实例
client = OCRClient("自己的x-ti-app-id", "自己的x-ti-secret-code")

# 读取图片文件
# 方式一：读取本地文件
with open("sample_02.pdf", "rb") as f:
file_content = f.read()
# 方式二：使用URL方式（需要将headers中的Content-Type改为'text/plain'）
# file_content = "https://example.com/path/to/your.pdf"

# 设置转换选项
options = dict(
apply_document_tree=1,
apply_merge=1,
catalog_details=1,
dpi=144,
formula_level=1,
get_excel=1,
get_image="objects",
markdown_details=1,
page_count=1000,
page_details=1,
paratext_mode="annotation",
parse_mode="scan",
table_flavor="html",
)

try:
response = client.recognize(file_content, options)

# 保存完整的JSON响应到result.json文件
with open("result.json", "w", encoding="utf-8") as f:
f.write(response)

# 解析JSON响应以提取markdown内容
json_response = json.loads(response)
if "result" in json_response and "markdown" in json_response["result"]:
markdown_content = json_response["result"]["markdown"]
with open("result.md", "w", encoding="utf-8") as f:
f.write(markdown_content)

print("Success! Result saved to result.json and result.md")
except Exception as e:
print(f"Error: {e}")

if __name__ == "__main__":
main()

**第三步：**解析自己的文件，我这里放的是sample_02.pdf文件，你可以自定义，代码注释详细，可以自定义配置。

一图了解整个流程

五、从解析到应用：下游集成实战

拿到解析结果之后，开发者最关心的问题是：怎么把它接入我的 RAG 系统或知识库？

以 TextIn xParse 的 JSON 输出为例，其结构化字段设计充分考虑了 AI 工作流的下游需求：

{"elements":[{"coordinates":[0.033155,0.010526,0.305348,0.010526,0.305348,0.014899,0.033155,0.014899],"element_id":"b0b28b7598462b1dc35628df9bd3086dbdfa0c4f15fea70bb9a4e328f07e137a","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"一、各部门Token用量与费用明细 核心报表","type":"Header"},{"coordinates":[0.031016,0.025749,0.960963,0.025101,0.962032,0.28,0.031551,0.28081],"element_id":"005d44743d4991491f5d4bada074ab5f22e4466730fc16d1bf18d30f8551f7d3","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"ref_element_id":"f7fe5462db30f641fafe7ab4f6c144c9310937325cb997d94fc507ca257ec08e"},"page_number":1,"sub_type":"borderless","text":"<table border=\\"1\\" ><tr>\\n<td rowspan=\\"2\\">部门</td>\\n<td rowspan=\\"2\\">二级部门</td>\\n<td rowspan=\\"2\\">负责人</td>\\n<td rowspan=\\"2\\">人数</td>\\n<td rowspan=\\"2\\">主要模型</td>\\n<td colspan=\\"2\\">Token用量</td>\\n<td rowspan=\\"2\\">费用（¥）</td>\\n<td rowspan=\\"2\\">预算（¥）</td>\\n<td rowspan=\\"2\\">执行率</td>\\n</tr><tr>\\n<td>输入Token</td>\\n<td>输出Token</td>\\n</tr><tr>\\n<td rowspan=\\"4\\">研发部</td>\\n<td>合计·95人</td>\\n<td></td>\\n<td>95</td>\\n<td>-</td>\\n<td>2,033.40亿</td>\\n<td>251.00亿</td>\\n<td>6,535,789.50</td>\\n<td>7,200,000.00</td>\\n<td>90.8%</td>\\n</tr><tr>\\n<td>AI平台组</td>\\n<td>张明远</td>\\n<td>45</td>\\n<td>GPT-4/Claude Opus／文心一言4.0</td>\\n<td>1,796.00亿</td>\\n<td>251.00亿</td>\\n<td>6,205,789.50</td>\\n<td>6,800,000.00</td>\\n<td>91.3%</td>\\n</tr><tr>\\n<td>基础架构组</td>\\n<td>李雪琴</td>\\n<td>32</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>770.00亿</td>\\n<td>115.00亿</td>\\n<td>2,670,000.00</td>\\n<td>3,000,000.00</td>\\n<td>89.0%</td>\\n</tr><tr>\\n<td>测试效能组</td>\\n<td>王建国</td>\\n<td>18</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>198.00亿</td>\\n<td>37.00亿</td>\\n<td>690,000.00</td>\\n<td>800,000.00</td>\\n<td>86.3%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">产品部</td>\\n<td>合计·50人</td>\\n<td></td>\\n<td>50</td>\\n<td>-</td>\\n<td>1,780.00亿</td>\\n<td>240.00亿</td>\\n<td>5,900,000.00</td>\\n<td>6,500,000.00</td>\\n<td>90.8%</td>\\n</tr><tr>\\n<td>智能产品线</td>\\n<td>赵雨婷</td>\\n<td>28</td>\\n<td>GPT-4/Claude Opus／通义千问</td>\\n<td>1,150.00亿</td>\\n<td>195.00亿</td>\\n<td>4,240,000.00</td>\\n<td>4,600,000.00</td>\\n<td>92.2%</td>\\n</tr><tr>\\n<td>数据产品线</td>\\n<td>孙浩然</td>\\n<td>22</td>\\n<td>GPT-4 / Claude Sonnet／文心一言4.0</td>\\n<td>740.00亿</td>\\n<td>115.00亿</td>\\n<td>2,330,000.00</td>\\n<td>2,600,000.00</td>\\n<td>89.6%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">市场部</td>\\n<td>合计·50人</td>\\n<td></td>\\n<td>50</td>\\n<td>-</td>\\n<td>1,195.00亿</td>\\n<td>206.00亿</td>\\n<td>5,453,000.00</td>\\n<td>5,800,000.00</td>\\n<td>94.0%</td>\\n</tr><tr>\\n<td>内容营销组</td>\\n<td>周思远</td>\\n<td>35</td>\\n<td>GPT-4/ Claude Opus/Claude Sonnet</td>\\n<td>950.00亿</td>\\n<td>257.00亿</td>\\n<td>5,450,000.00</td>\\n<td>5,800,000.00</td>\\n<td>94.0%</td>\\n</tr><tr>\\n<td>品牌设计组</td>\\n<td>吴佳琪</td>\\n<td>15</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>185.00亿</td>\\n<td>84.00亿</td>\\n<td>1,023,000.00</td>\\n<td>1,200,000.00</td>\\n<td>85.3%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">运营部</td>\\n<td>合计·65人</td>\\n<td></td>\\n<td>65</td>\\n<td>-</td>\\n<td>1,990.00亿</td>\\n<td>485.00亿</td>\\n<td>10,800,000.00</td>\\n<td>12,000,000.00</td>\\n<td>90.0%</td>\\n</tr><tr>\\n<td>用户运营组</td>\\n<td>郑子涵</td>\\n<td>40</td>\\n<td>GPT-4/Claude Opus/Claude Sonnet／文心一言4.0</td>\\n<td>1,990.00亿</td>\\n<td>785.00亿</td>\\n<td>10,550,000.00</td>\\n<td>11,500,000.00</td>\\n<td>91.7%</td>\\n</tr><tr>\\n<td>数据运营组</td>\\n<td>陆雨萱</td>\\n<td>25</td>\\n<td>GPT-4/Claude Opus／通义千问</td>\\n<td>880.00亿</td>\\n<td>425.00亿</td>\\n<td>5,010,000.00</td>\\n<td>5,500,000.00</td>\\n<td>91.1%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">数据科学部</td>\\n<td>合计·58人</td>\\n<td></td>\\n<td>58</td>\\n<td>-</td>\\n<td>2,940.00亿</td>\\n<td>372.00亿</td>\\n<td>14,814,000.00</td>\\n<td>16,000,000.00</td>\\n<td>92.6%</td>\\n</tr><tr>\\n<td>算法研究组</td>\\n<td>钱伟豪</td>\\n<td>38</td>\\n<td>GPT-4 / Claude Opus / Claude Sonnet／文心一言4.0／通义千问</td>\\n<td>2,950.00亿</td>\\n<td>467.00亿</td>\\n<td>14,810,000.00</td>\\n<td>16,000,000.00</td>\\n<td>92.6%</td>\\n</tr><tr>\\n<td>数据工程组</td>\\n<td>陈俊杰</td>\\n<td>20</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>248.00亿</td>\\n<td>70.00亿</td>\\n<td>1,134,000.00</td>\\n<td>1,300,000.00</td>\\n<td>87.2%</td>\\n</tr><tr>\\n<td rowspan=\\"2\\">财务部</td>\\n<td>合计·12人</td>\\n<td></td>\\n<td>12</td>\\n<td>-</td>\\n<td>90.00亿</td>\\n<td>27.50亿</td>\\n<td>434,000.00</td>\\n<td>500,000.00</td>\\n<td>86.8%</td>\\n</tr><tr>\\n<td>财务分析组</td>\\n<td>林妙音</td>\\n<td>12</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>90.00亿</td>\\n<td>27.50亿</td>\\n<td>434,000.00</td>\\n<td>500,000.00</td>\\n<td>86.8%</td>\\n</tr><tr>\\n<td>集团合计</td>\\n<td></td>\\n<td></td>\\n<td></td>\\n<td>-</td>\\n<td>10,028.40亿</td>\\n<td>1,906.50亿</td>\\n<td>¥43,936,789.50</td>\\n<td>¥48,000,000.00</td>\\n<td>91.5%</td>\\n</tr></table>","type":"Table"},{"coordinates":[0.034225,0.300567,0.213369,0.300567,0.213369,0.304939,0.034225,0.304939],"element_id":"f7fe5462db30f641fafe7ab4f6c144c9310937325cb997d94fc507ca257ec08e","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"**二、各模型用量与费用对比**","type":"TableCaption"},{"coordinates":[0.031551,0.313684,0.5,0.313684,0.49893,0.410364,0.029412,0.410364],"element_id":"913ee7e62fc92de95e4d7c80b9a84ded6ef77ceb22c86649a7f556f8cad5dd45","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"ref_element_id":"74c3526f3d69b8129067aaf8df191f4fdba8d7e9bab184723712ce93b7ed92a1"},"page_number":1,"sub_type":"borderless","text":"<table border=\\"1\\" ><tr>\\n<td>模型</td>\\n<td>调用次数</td>\\n<td>Token小计</td>\\n<td>总费用（¥）</td>\\n<td>干Token均价（¥）</td>\\n<td>费用占比</td>\\n</tr><tr>\\n<td>GPT-4</td>\\n<td>15,680,000</td>\\n<td>4,720.00亿</td>\\n<td>21,278,000.00</td>\\n<td>0.045</td>\\n<td>43.2%</td>\\n</tr><tr>\\n<td>ClaudeOpus</td>\\n<td>8,950,000</td>\\n<td>3,515.50亿</td>\\n<td>15,645,000.00</td>\\n<td>0.044</td>\\n<td>31.8%</td>\\n</tr><tr>\\n<td>ClaudeSonnet</td>\\n<td>8,720,000</td>\\n<td>2,034.00亿</td>\\n<td>6,885,000.00</td>\\n<td>0.034</td>\\n<td>14.0%</td>\\n</tr><tr>\\n<td>文心一言4.0</td>\\n<td>4,670,000</td>\\n<td>1,267.20亿</td>\\n<td>3,310,000.00</td>\\n<td>0.026</td>\\n<td>6.7%</td>\\n</tr><tr>\\n<td>通义千问</td>\\n<td>2,960,000</td>\\n<td>724.50亿</td>\\n<td>1,570,000.00</td>\\n<td>0.022</td>\\n<td>3.2%</td>\\n</tr><tr>\\n<td>合计</td>\\n<td>40,980,000</td>\\n<td>12,261.20亿</td>\\n<td>48,688,000.00</td>\\n<td>-</td>\\n<td>100%</td>\\n</tr></table>","type":"Table"},{"coordinates":[0.568449,0.300567,0.728877,0.300567,0.728877,0.305101,0.568449,0.305101],"element_id":"74c3526f3d69b8129067aaf8df191f4fdba8d7e9bab184723712ce93b7ed92a1","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"**三、月度趋势（2025年）**","type":"TableCaption"},{"coordinates":[0.565775,0.313036,0.962567,0.313198,0.963102,0.443725,0.565775,0.443725],"element_id":"151a7f0d9eecc434cbe14a1b79375060f837ec40226c689e12e885d3692ebeba","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"sub_type":"borderless","text":"<table border=\\"1\\" ><tr>\\n<td>月份</td>\\n<td>调用次数</td>\\n<td>Token总量</td>\\n<td>费用（¥）</td>\\n<td>环比（费用）</td>\\n</tr><tr>\\n<td>1月</td>\\n<td>2,575,000</td>\\n<td>485.00亿</td>\\n<td>2,722,000.00</td>\\n<td>-</td>\\n</tr><tr>\\n<td>2月</td>\\n<td>2,472,000</td>\\n<td>464.00亿</td>\\n<td>2,605,000.00</td>\\n<td>-4.3%<img src=\\"https://web-api.textin.com/ocr_image/external/2a10b2c1397dab4b.jpg\\"></td>\\n</tr><tr>\\n<td>3月</td>\\n<td>3,038,000</td>\\n<td>572.00亿</td>\\n<td>3,205,000.00</td>\\n<td>+23.0%</td>\\n</tr><tr>\\n<td>4月</td>\\n<td>3,263,000</td>\\n<td>614.00亿</td>\\n<td>3,446,000.00</td>\\n<td>+7.5%</td>\\n</tr><tr>\\n<td>5月</td>\\n<td>3,464,000</td>\\n<td>652.00亿</td>\\n<td>3,659,000.00</td>\\n<td>+6.2%</td>\\n</tr><tr>\\n<td>6月</td>\\n<td>3,630,000</td>\\n<td>683.00亿</td>\\n<td>3,839,000.00</td>\\n<td>+4.9%</td>\\n</tr><tr>\\n<td>7月</td>\\n<td>3,538,000</td>\\n<td>665.00亿</td>\\n<td>3,733,000.00</td>\\n<td>-2.8%</td>\\n</tr><tr>\\n<td>8月</td>\\n<td>3,721,000</td>\\n<td>700.00亿</td>\\n<td>3,936,000.00</td>\\n<td>+5.4%</td>\\n</tr><tr>\\n<td>9月</td>\\n<td>3,646,000</td>\\n<td>685.00亿</td>\\n<td>3,850,000.00</td>\\n<td>-2.2%</td>\\n</tr><tr>\\n<td>10月</td>\\n<td>3,839,000</td>\\n<td>721.00亿</td>\\n<td>4,054,000.00</td>\\n<td>+5.3%</td>\\n</tr><tr>\\n<td>11月</td>\\n<td>3,928,000</td>\\n<td>739.00亿</td>\\n<td>4,153,000.00</td>\\n<td>+2.4%</td>\\n</tr><tr>\\n<td>12月</td>\\n<td>4,146,000</td>\\n<td>780.00亿</td>\\n<td>4,388,000.00</td>\\n<td>+5.7%</td>\\n</tr><tr>\\n<td>累计</td>\\n<td>42,260,000</td>\\n<td>7,760.00亿</td>\\n<td>43,590,000.00</td>\\n<td>-</td>\\n</tr></table>","type":"Table"},{"coordinates":[0.568984,0.446802,0.948663,0.447449,0.948663,0.495223,0.568984,0.494575],"element_id":"3718ef78786ac3fe0afc9a7ea16d69c0688a0739cfa32f5448e8f0dfa32e295b","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":299,"is_continuation":false,"width":710},"page_number":1,"sub_type":"chart","text":"440385月度费用366384373394345321400272261(A)350旺邮3002501月2月3月4月5月6月7月8月9月10月11月12月","type":"Image"},{"coordinates":[0.03369,0.521943,0.201604,0.521943,0.201604,0.52664,0.03369,0.52664],"element_id":"fa37bc71e569e9fa6e29d76627af96d280c2b13bfb548fdd20af7a1765e919db","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"**四、Top 10用量用户排行**","type":"TableCaption"},{"coordinates":[0.026203,0.534413,0.958289,0.534413,0.959893,0.647287,0.027807,0.647287],"element_id":"27293311abf02020ae64ebec82104dfd5d9b55a71a81cf195a3e3f88a06fe83d","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"ref_element_id":"fa37bc71e569e9fa6e29d76627af96d280c2b13bfb548fdd20af7a1765e919db"},"page_number":1,"sub_type":"borderless","text":"<table border=\\"1\\" ><tr>\\n<td>排名</td>\\n<td>姓名</td>\\n<td>所属部门／二级部门</td>\\n<td>API调用次数</td>\\n<td>Token总量</td>\\n<td>费用（¥）</td>\\n<td>人均费用排名</td>\\n</tr><tr>\\n<td>1<img src=\\"https://web-api.textin.com/ocr_image/external/ba7f3d967ae7e1e0.jpg\\"></td>\\n<td>钱伟豪</td>\\n<td>数据科学部／算法研究组</td>\\n<td>2,450,000</td>\\n<td>485.00亿</td>\\n<td>1,890,000.00</td>\\n<td>1</td>\\n</tr><tr>\\n<td>2<img src=\\"https://web-api.textin.com/ocr_image/external/53bc114224d67844.jpg\\"></td>\\n<td>郑子涵</td>\\n<td>运营部／用户运营组</td>\\n<td>2,280,000</td>\\n<td>420.00亿</td>\\n<td>1,750,000.00</td>\\n<td>2</td>\\n</tr><tr>\\n<td>3<img src=\\"https://web-api.textin.com/ocr_image/external/43d1eafd23a09124.jpg\\"></td>\\n<td>张明远</td>\\n<td>研发部／AI平台组</td>\\n<td>2,150,000</td>\\n<td>395.00亿</td>\\n<td>1,620,000.00</td>\\n<td>3</td>\\n</tr><tr>\\n<td>4<img src=\\"https://web-api.textin.com/ocr_image/external/e18b5369f3912613.jpg\\"></td>\\n<td>陆雨萱</td>\\n<td>运营部／数据运营组</td>\\n<td>1,980,000</td>\\n<td>370.00亿</td>\\n<td>1,480,000.00</td>\\n<td>4</td>\\n</tr><tr>\\n<td>5<img src=\\"https://web-api.textin.com/ocr_image/external/48f179722aeb7feb.jpg\\"></td>\\n<td>周思远</td>\\n<td>市场部／内容营销组</td>\\n<td>1,850,000</td>\\n<td>340.00亿</td>\\n<td>1,420,000.00</td>\\n<td>5</td>\\n</tr><tr>\\n<td>6<img src=\\"https://web-api.textin.com/ocr_image/external/ea746cadc19fece9.jpg\\"></td>\\n<td>赵雨婷</td>\\n<td>产品部／智能产品线</td>\\n<td>1,720,000</td>\\n<td>315.00亿</td>\\n<td>1,280,000.00</td>\\n<td>6</td>\\n</tr><tr>\\n<td>7<img src=\\"https://web-api.textin.com/ocr_image/external/b791b3f37b98a5d8.jpg\\"></td>\\n<td>孙浩然</td>\\n<td>产品部／数据产品线</td>\\n<td>1,580,000</td>\\n<td>290.00亿</td>\\n<td>1,180,000.00</td>\\n<td>7</td>\\n</tr><tr>\\n<td>8<img src=\\"https://web-api.textin.com/ocr_image/external/0c046b802aee1247.jpg\\"></td>\\n<td>李雪琴</td>\\n<td>研发部／基础架构组</td>\\n<td>1,350,000</td>\\n<td>250.00亿</td>\\n<td>1,050,000.00</td>\\n<td>8</td>\\n</tr><tr>\\n<td>9<img src=\\"https://web-api.textin.com/ocr_image/external/0aa4be2f11a1a87a.jpg\\"></td>\\n<td>吴佳琪</td>\\n<td>市场部／品牌设计组</td>\\n<td>980,000</td>\\n<td>185.00亿</td>\\n<td>780,000.00</td>\\n<td>9</td>\\n</tr><tr>\\n<td>10<img src=\\"https://web-api.textin.com/ocr_image/external/ed61a7dd2dacb18a.jpg\\"></td>\\n<td>王建国</td>\\n<td>研发部／测试效能组</td>\\n<td>850,000</td>\\n<td>160.00亿</td>\\n<td>680,000.00</td>\\n<td>11</td>\\n</tr></table>","type":"Table"},{"coordinates":[0.03369,0.648907,0.467914,0.648907,0.467914,0.652146,0.03369,0.652146],"element_id":"6be7c91a4d9f873327966ec3f0f878de7f6002c111cf116bc7ad423f5956cb71","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"注：王建国人均费用排名第11位，因其组内人数较少（18人），人均费用略高于部分排名靠前的用户。","type":"NarrativeText"},{"coordinates":[0.035294,0.673522,0.186096,0.673522,0.186096,0.678057,0.035294,0.678057],"element_id":"6ab79c8789460b2b36aabcec10b91e869d18270409a3e5c1835e95858e4f4ebd","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"**五、项目级用量 Top 10**","type":"TableCaption"},{"coordinates":[0.026738,0.688097,0.963636,0.687126,0.964706,0.795789,0.027273,0.796761],"element_id":"ee9df3e9367c280e8e7f8799dc3d4a2572d455c8b11e8a945cce043f371e4702","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"ref_element_id":"6ab79c8789460b2b36aabcec10b91e869d18270409a3e5c1835e95858e4f4ebd"},"page_number":1,"sub_type":"borderless","text":"<table border=\\"1\\" ><tr>\\n<td>排名</td>\\n<td>项目名称</td>\\n<td>所属部门</td>\\n<td>负责人</td>\\n<td>API调用</td>\\n<td>Token总量</td>\\n<td>费用（¥）</td>\\n<td>主要使用模型</td>\\n</tr><tr>\\n<td><img src=\\"https://web-api.textin.com/ocr_image/external/06956064d4cf02d3.jpg\\"></td>\\n<td>智能客服</td>\\n<td>研发部／AI平台组</td>\\n<td>张明远</td>\\n<td>5,800,000</td>\\n<td>892.00亿</td>\\n<td>3,780,000.00</td>\\n<td>GPT-4/ Claude Opus</td>\\n</tr><tr>\\n<td>2<img src=\\"https://web-api.textin.com/ocr_image/external/f3c6386ca77d6467.jpg\\"></td>\\n<td>大模型微调</td>\\n<td>数据科学部／算法研究组</td>\\n<td>钱伟豪</td>\\n<td>5,200,000</td>\\n<td>785.00亿</td>\\n<td>3,450,000.00</td>\\n<td>Claude Opus/GPT-4</td>\\n</tr><tr>\\n<td>3<img src=\\"https://web-api.textin.com/ocr_image/external/cb6fbea2a2a9ede1.jpg\\"></td>\\n<td>用户画像</td>\\n<td>运营部／用户运营组</td>\\n<td>郑子涵</td>\\n<td>4,800,000</td>\\n<td>720.00亿</td>\\n<td>3,120,000.00</td>\\n<td>GPT-4 / Claude Sonnet</td>\\n</tr><tr>\\n<td>4<img src=\\"https://web-api.textin.com/ocr_image/external/dc31a7ad512f72b6.jpg\\"></td>\\n<td>AI搜索</td>\\n<td>产品部／智能产品线</td>\\n<td>赵雨婷</td>\\n<td>4,200,000</td>\\n<td>620.00亿</td>\\n<td>2,680,000.00</td>\\n<td>GPT-4／通义千问</td>\\n</tr><tr>\\n<td>5<img src=\\"https://web-api.textin.com/ocr_image/external/0f1482ed0eead7af.jpg\\"></td>\\n<td>内容生成</td>\\n<td>市场部／内容营销组</td>\\n<td>周思远</td>\\n<td>4,000,000</td>\\n<td>585.00亿</td>\\n<td>2,540,000.00</td>\\n<td>GPT-4/Claude Sonnet</td>\\n</tr><tr>\\n<td>6</td>\\n<td>精准推送</td>\\n<td>运营部／用户运营组</td>\\n<td>郑子涵</td>\\n<td>3,600,000</td>\\n<td>530.00亿</td>\\n<td>2,300,000.00</td>\\n<td>Claude Opus / GPT-4</td>\\n</tr><tr>\\n<td>7</td>\\n<td>文档理解</td>\\n<td>数据科学部／算法研究组</td>\\n<td>钱伟豪</td>\\n<td>3,400,000</td>\\n<td>500.00亿</td>\\n<td>2,180,000.00</td>\\n<td>GPT-4/Claude Sonnet</td>\\n</tr><tr>\\n<td>8<img src=\\"https://web-api.textin.com/ocr_image/external/f2c2ba459ea83fed.jpg\\"></td>\\n<td>代码助手</td>\\n<td>研发部／AI平台组</td>\\n<td>张明远</td>\\n<td>3,200,000</td>\\n<td>475.00亿</td>\\n<td>2,050,000.00</td>\\n<td>Claude Opus/GPT-4</td>\\n</tr><tr>\\n<td>9<img src=\\"https://web-api.textin.com/ocr_image/external/b777c77519a0fb55.jpg\\"></td>\\n<td>A/B测试平台</td>\\n<td>运营部／数据运营组</td>\\n<td>陆雨萱</td>\\n<td>2,900,000</td>\\n<td>430.00亿</td>\\n<td>1,860,000.00</td>\\n<td>GPT-4／通义千问</td>\\n</tr><tr>\\n<td>10<img src=\\"https://web-api.textin.com/ocr_image/external/bd6fc7e64cf682f4.jpg\\"></td>\\n<td>智能问答</td>\\n<td>产品部／智能产品线</td>\\n<td>赵雨婷</td>\\n<td>2,700,000</td>\\n<td>400.00亿</td>\\n<td>1,730,000.00</td>\\n<td>GPT-4／文心一言4.0</td>\\n</tr></table>","type":"Table"},{"coordinates":[0.03369,0.817328,0.19893,0.817328,0.19893,0.821862,0.03369,0.821862],"element_id":"59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf","metadata":{"category_depth":0,"children_ids":["0d361f4d825cf1d411087be19d366b2a2cd5962a23ca9e85c772196ff835238b","4fc3d5bfc0b7071ebe9802e674ced047934dcbbf4f5ba71163cc5d76e53355f4","0ed120e82f9d23db44283751a04bfe7f8d9ee7ae4d2bb97086eac6a1f2f72028","45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false},"page_number":1,"text":"六、异常告警与优化建议","type":"Title"},{"coordinates":[0.054011,0.835789,0.074332,0.835789,0.074332,0.841457,0.054011,0.841457],"element_id":"0d361f4d825cf1d411087be19d366b2a2cd5962a23ca9e85c772196ff835238b","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":35,"is_continuation":false,"parent_id":"59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf","width":38},"page_number":1,"text":"","type":"Image"},{"coordinates":[0.080214,0.836113,0.249733,0.836113,0.249733,0.839676,0.080214,0.839676],"element_id":"4fc3d5bfc0b7071ebe9802e674ced047934dcbbf4f5ba71163cc5d76e53355f4","metadata":{"category_depth":1,"children_ids":["f85a431352d4bbaee1f2703ebb3ce1facd38e95ddbdae088cc2588bb1196c982","d890fc45661c5a8c08ea3ac8ee007eeb8842d1371b47f005031cb736130c80f4"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf"},"page_number":1,"text":"内容营销组12月API调用激增","type":"Title"},{"coordinates":[0.080214,0.842591,0.793583,0.842591,0.793583,0.84664,0.080214,0.84664],"element_id":"f85a431352d4bbaee1f2703ebb3ce1facd38e95ddbdae088cc2588bb1196c982","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"4fc3d5bfc0b7071ebe9802e674ced047934dcbbf4f5ba71163cc5d76e53355f4"},"page_number":1,"text":"12月份调用720,000次，较11月（680,000次）增长5.9%，较9月（630,000次）增长14.3％。建议确认是否存在非授权批量调用。","type":"NarrativeText"},{"coordinates":[0.054545,0.857814,0.074332,0.857814,0.074332,0.863968,0.054545,0.863968],"element_id":"d890fc45661c5a8c08ea3ac8ee007eeb8842d1371b47f005031cb736130c80f4","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":38,"is_continuation":false,"parent_id":"4fc3d5bfc0b7071ebe9802e674ced047934dcbbf4f5ba71163cc5d76e53355f4","width":37},"page_number":1,"text":"","type":"Image"},{"coordinates":[0.05508,0.858785,0.32139,0.858785,0.32139,0.86251,0.05508,0.86251],"element_id":"0ed120e82f9d23db44283751a04bfe7f8d9ee7ae4d2bb97086eac6a1f2f72028","metadata":{"category_depth":1,"children_ids":["6841538746fc9f3da46a66c24cd07a07c5c0e6558a4075bcd0ce148a98b0f173","40507e4a4de837fac52069b3cd8052b958d0521b70dc11db8a4fe7b2fea5758f"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf"},"page_number":1,"text":"算法研究组 Claude Opus 费用严重超预算","type":"Title"},{"coordinates":[0.079144,0.865101,0.942781,0.865101,0.942781,0.875466,0.079144,0.875466],"element_id":"6841538746fc9f3da46a66c24cd07a07c5c0e6558a4075bcd0ce148a98b0f173","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"0ed120e82f9d23db44283751a04bfe7f8d9ee7ae4d2bb97086eac6a1f2f72028"},"page_number":1,"text":"Claude Opus全年费用612万元，占部门总费用的41.3%，较年初预算超支12％。建议评估是否可将部分任务迁移至Claude Sonnet以降低费用。按干Token 差价0.01元计算，若迁移30％的Opus调用量，预计年节省费用约110万元。","type":"NarrativeText"},{"coordinates":[0.054545,0.886964,0.073262,0.886964,0.073262,0.892794,0.054011,0.892794],"element_id":"40507e4a4de837fac52069b3cd8052b958d0521b70dc11db8a4fe7b2fea5758f","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":36,"is_continuation":false,"parent_id":"0ed120e82f9d23db44283751a04bfe7f8d9ee7ae4d2bb97086eac6a1f2f72028","width":35},"page_number":1,"text":"","type":"Image"},{"coordinates":[0.079144,0.887611,0.301604,0.887611,0.301604,0.891498,0.079144,0.891498],"element_id":"45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c","metadata":{"category_depth":1,"children_ids":["d8dbc7a9e5eb026f2dffa9644d6f105a2dfee671e900b3a7cb4523244e0f7020","41096bd2e917c5105652fad5b6d174762ca9d214f08f430b9857405102eb53bd","cbf50cca607f8489d297a89badd62f1c00bdc880c7728fb13cb80882bb4cb285","0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf"},"page_number":1,"text":"国产模型占比偏低，成本优化空间显著","type":"Title"},{"coordinates":[0.079144,0.894089,0.949198,0.894089,0.949198,0.904453,0.079144,0.904453],"element_id":"d8dbc7a9e5eb026f2dffa9644d6f105a2dfee671e900b3a7cb4523244e0f7020","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c"},"page_number":1,"text":"文心一言4.0千Token均价0.026元，通义千问0.022元，均显著低于GPT-4(0.045元）和Claude Opus(0.044元）。当前国产模型调用次数占比仅18.3%，费用占比仅9.9％。用户运营组文心一言调用占比14.6%，建议全集团层面增加国产模型使用比例以控制成本。","type":"NarrativeText"},{"coordinates":[0.054011,0.915951,0.073797,0.915789,0.073797,0.921619,0.054011,0.921619],"element_id":"41096bd2e917c5105652fad5b6d174762ca9d214f08f430b9857405102eb53bd","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":35,"is_continuation":false,"parent_id":"45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c","width":37},"page_number":1,"text":"","type":"Image"},{"coordinates":[0.054545,0.916275,0.25508,0.916275,0.25508,0.920162,0.054545,0.920162],"element_id":"cbf50cca607f8489d297a89badd62f1c00bdc880c7728fb13cb80882bb4cb285","metadata":{"category_depth":2,"children_ids":["8849f6dbfbe793665ebdb3e6e307d1d8797c264b9ebe06a404316ed8c44890dd","4a6dbe6256f876cf819f1d197fa3470a763b1555bdb625aff77c81a847f29333"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c"},"page_number":1,"text":"数据工程组Q4调用量平稳回落","type":"Title"},{"coordinates":[0.079144,0.923077,0.73262,0.923077,0.73262,0.927126,0.079144,0.927126],"element_id":"8849f6dbfbe793665ebdb3e6e307d1d8797c264b9ebe06a404316ed8c44890dd","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"cbf50cca607f8489d297a89badd62f1c00bdc880c7728fb13cb80882bb4cb285"},"page_number":1,"text":"Q4月度调用量稳定在12-13万次，较Q2巅峰（11.5万次）有所回落，H2整体趋势平稳，属正常波动，无需特殊关注。","type":"NarrativeText"},{"coordinates":[0.05508,0.938947,0.073262,0.938785,0.073797,0.944453,0.05508,0.944453],"element_id":"4a6dbe6256f876cf819f1d197fa3470a763b1555bdb625aff77c81a847f29333","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"height":34,"is_continuation":false,"parent_id":"cbf50cca607f8489d297a89badd62f1c00bdc880c7728fb13cb80882bb4cb285","width":34},"page_number":1,"text":"","type":"Image"},{"coordinates":[0.080214,0.938947,0.264706,0.938947,0.264706,0.942996,0.080214,0.942996],"element_id":"0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce","metadata":{"category_depth":2,"children_ids":["4a922b435a499e99243cb96ed4fbbac91c8904ac6cf85f7ab6e418f808e737cf","d1a7efcef9dc649c3fcda506755c4144406831abe0d5dd0f423b6d5ba4dc01dc","05ae9dc9fe33dadf39ca2c13203e369a804bd771a0a84782055e86b828cc145e"],"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c"},"page_number":1,"text":"品牌设计组人均费用全集团最高","type":"Title"},{"coordinates":[0.080214,0.945101,0.930481,0.945101,0.930481,0.955789,0.080214,0.955789],"element_id":"4a922b435a499e99243cb96ed4fbbac91c8904ac6cf85f7ab6e418f808e737cf","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce"},"page_number":1,"text":"品牌设计组15人，人均费用68,200元／年，为全公司最高（全公司均值29,000元／年，为2.35倍）。建议审查是否存在个人账号混用或多项目共用同-API Key的情况。","type":"NarrativeText"},{"coordinates":[0.334759,0.981377,0.664171,0.981377,0.664171,0.984453,0.334759,0.984453],"element_id":"d1a7efcef9dc649c3fcda506755c4144406831abe0d5dd0f423b6d5ba4dc01dc","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce"},"page_number":1,"text":"制表日期：2026-01-10｜报表版本：v2.3｜下次报表生成日期：2026-02-05","type":"NarrativeText"},{"coordinates":[0.360428,0.986073,0.637433,0.986073,0.637433,0.989312,0.360428,0.989312],"element_id":"05ae9dc9fe33dadf39ca2c13203e369a804bd771a0a84782055e86b828cc145e","metadata":{"category_depth":-1,"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"is_continuation":false,"parent_id":"0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce"},"page_number":1,"text":"原始数据来源于集团AI网关日志系统｜本报表仅供内部管理使用","type":"NarrativeText"}],"file_id":"49a0ab97a3c14bf18b88b8d49cbae812","history_info":{"cloud_ocr":0,"ctime":"1781096956","id":5493532,"img_name":"某某公司AI用量报表.png","img_uri":"https://web-api.textin.com/image/download?filename=9325751c5ddd4efba30783a88b.png","ocr_status":1,"status":0,"thumbnail":"","utime":"1781096956"},"job_id":"9891a8e8b7cb4d1f88799b76527e0fc5","markdown":"<!– **一、各部门Token用量与费用明细** 核心报表 –>\\n\\n<table border=\\"1\\" ><tr>\\n<td rowspan=\\"2\\">部门</td>\\n<td rowspan=\\"2\\">二级部门</td>\\n<td rowspan=\\"2\\">负责人</td>\\n<td rowspan=\\"2\\">人数</td>\\n<td rowspan=\\"2\\">主要模型</td>\\n<td colspan=\\"2\\">Token 用量</td>\\n<td rowspan=\\"2\\">费用（¥）</td>\\n<td rowspan=\\"2\\">预算（¥）</td>\\n<td rowspan=\\"2\\">执行率</td>\\n</tr><tr>\\n<td>输入Token</td>\\n<td>输出Token</td>\\n</tr><tr>\\n<td rowspan=\\"4\\">研发部</td>\\n<td>合计·95 人</td>\\n<td></td>\\n<td>95</td>\\n<td>-</td>\\n<td>2,033.40 亿</td>\\n<td>251.00 亿</td>\\n<td>6,535,789.50</td>\\n<td>7,200,000.00</td>\\n<td>90.8%</td>\\n</tr><tr>\\n<td>AI 平台组</td>\\n<td>张明远</td>\\n<td>45</td>\\n<td>GPT-4/Claude Opus／文心一言4.0</td>\\n<td>1,796.00 亿</td>\\n<td>251.00 亿</td>\\n<td>6,205,789.50</td>\\n<td>6,800,000.00</td>\\n<td>91.3%</td>\\n</tr><tr>\\n<td>基础架构<br>组</td>\\n<td>李雪琴</td>\\n<td>32</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>770.00 亿</td>\\n<td>115.00 亿</td>\\n<td>2,670,000.00</td>\\n<td>3,000,000.00</td>\\n<td>89.0%</td>\\n</tr><tr>\\n<td>测试效能<br>组</td>\\n<td>王建国</td>\\n<td>18</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>198.00 亿</td>\\n<td>37.00 亿</td>\\n<td>690,000.00</td>\\n<td>800,000.00</td>\\n<td>86.3%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">产品部</td>\\n<td>合计·50 人</td>\\n<td></td>\\n<td>50</td>\\n<td>-</td>\\n<td>1,780.00 亿</td>\\n<td>240.00 亿</td>\\n<td>5,900,000.00</td>\\n<td>6,500,000.00</td>\\n<td>90.8%</td>\\n</tr><tr>\\n<td>智能产品<br>线</td>\\n<td>赵雨婷</td>\\n<td>28</td>\\n<td>GPT-4/Claude Opus／通义千问</td>\\n<td>1,150.00 亿</td>\\n<td>195.00 亿</td>\\n<td>4,240,000.00</td>\\n<td>4,600,000.00</td>\\n<td>92.2%</td>\\n</tr><tr>\\n<td>数据产品<br>线</td>\\n<td>孙浩然</td>\\n<td>22</td>\\n<td>GPT-4 / Claude Sonnet／文心一言4.0</td>\\n<td>740.00 亿</td>\\n<td>115.00 亿</td>\\n<td>2,330,000.00</td>\\n<td>2,600,000.00</td>\\n<td>89.6%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">市场部</td>\\n<td>合计·50 人</td>\\n<td></td>\\n<td>50</td>\\n<td>-</td>\\n<td>1,195.00 亿</td>\\n<td>206.00 亿</td>\\n<td>5,453,000.00</td>\\n<td>5,800,000.00</td>\\n<td>94.0%</td>\\n</tr><tr>\\n<td>内容营销<br>组</td>\\n<td>周思远</td>\\n<td>35</td>\\n<td>GPT-4/ Claude Opus/Claude Sonnet</td>\\n<td>950.00 亿</td>\\n<td>257.00 亿</td>\\n<td>5,450,000.00</td>\\n<td>5,800,000.00</td>\\n<td>94.0%</td>\\n</tr><tr>\\n<td>品牌设计<br>组</td>\\n<td>吴佳琪</td>\\n<td>15</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>185.00 亿</td>\\n<td>84.00 亿</td>\\n<td>1,023,000.00</td>\\n<td>1,200,000.00</td>\\n<td>85.3%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">运营部</td>\\n<td>合计·65 人</td>\\n<td></td>\\n<td>65</td>\\n<td>-</td>\\n<td>1,990.00 亿</td>\\n<td>485.00 亿</td>\\n<td>10,800,000.00</td>\\n<td>12,000,000.00</td>\\n<td>90.0%</td>\\n</tr><tr>\\n<td>用户运营<br>组</td>\\n<td>郑子涵</td>\\n<td>40</td>\\n<td>GPT-4/Claude Opus/Claude Sonnet／文心一言4.0</td>\\n<td>1,990.00 亿</td>\\n<td>785.00 亿</td>\\n<td>10,550,000.00</td>\\n<td>11,500,000.00</td>\\n<td>91.7%</td>\\n</tr><tr>\\n<td>数据运营<br>组</td>\\n<td>陆雨萱</td>\\n<td>25</td>\\n<td>GPT-4/Claude Opus／通义千问</td>\\n<td>880.00 亿</td>\\n<td>425.00 亿</td>\\n<td>5,010,000.00</td>\\n<td>5,500,000.00</td>\\n<td>91.1%</td>\\n</tr><tr>\\n<td rowspan=\\"3\\">数据科学<br>部</td>\\n<td>合计·58 人</td>\\n<td></td>\\n<td>58</td>\\n<td>-</td>\\n<td>2,940.00 亿</td>\\n<td>372.00 亿</td>\\n<td>14,814,000.00</td>\\n<td>16,000,000.00</td>\\n<td>92.6%</td>\\n</tr><tr>\\n<td>算法研究<br>组</td>\\n<td>钱伟豪</td>\\n<td>38</td>\\n<td>GPT-4 / Claude Opus / Claude Sonnet／文心一言4.0／<br>通义千问</td>\\n<td>2,950.00 亿</td>\\n<td>467.00 亿</td>\\n<td>14,810,000.00</td>\\n<td>16,000,000.00</td>\\n<td>92.6%</td>\\n</tr><tr>\\n<td>数据工程<br>组</td>\\n<td>陈俊杰</td>\\n<td>20</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>248.00 亿</td>\\n<td>70.00 亿</td>\\n<td>1,134,000.00</td>\\n<td>1,300,000.00</td>\\n<td>87.2%</td>\\n</tr><tr>\\n<td rowspan=\\"2\\">财务部</td>\\n<td>合计·12 人</td>\\n<td></td>\\n<td>12</td>\\n<td>-</td>\\n<td>90.00 亿</td>\\n<td>27.50 亿</td>\\n<td>434,000.00</td>\\n<td>500,000.00</td>\\n<td>86.8%</td>\\n</tr><tr>\\n<td>财务分析<br>组</td>\\n<td>林妙音</td>\\n<td>12</td>\\n<td>GPT-4/Claude Sonnet</td>\\n<td>90.00 亿</td>\\n<td>27.50 亿</td>\\n<td>434,000.00</td>\\n<td>500,000.00</td>\\n<td>86.8%</td>\\n</tr><tr>\\n<td>集团合计</td>\\n<td></td>\\n<td></td>\\n<td></td>\\n<td>-</td>\\n<td>10,028.40<br>亿</td>\\n<td>1,906.50<br>亿</td>\\n<td>¥43,936,789.50</td>\\n<td>¥48,000,000.00</td>\\n<td>91.5%</td>\\n</tr></table>\\n\\n**二、各模型用量与费用对比**\\n\\n<table border=\\"1\\" ><tr>\\n<td>模型</td>\\n<td>调用次数</td>\\n<td>Token 小计</td>\\n<td>总费用（¥）</td>\\n<td>干Token 均价（¥）</td>\\n<td>费用占比</td>\\n</tr><tr>\\n<td>GPT-4</td>\\n<td>15,680,000</td>\\n<td>4,720.00 亿</td>\\n<td>21,278,000.00</td>\\n<td>0.045</td>\\n<td>43.2%</td>\\n</tr><tr>\\n<td>Claude Opus</td>\\n<td>8,950,000</td>\\n<td>3,515.50 亿</td>\\n<td>15,645,000.00</td>\\n<td>0.044</td>\\n<td>31.8%</td>\\n</tr><tr>\\n<td>Claude Sonnet</td>\\n<td>8,720,000</td>\\n<td>2,034.00 亿</td>\\n<td>6,885,000.00</td>\\n<td>0.034</td>\\n<td>14.0%</td>\\n</tr><tr>\\n<td>文心一言4.0</td>\\n<td>4,670,000</td>\\n<td>1,267.20 亿</td>\\n<td>3,310,000.00</td>\\n<td>0.026</td>\\n<td>6.7%</td>\\n</tr><tr>\\n<td>通义千<br>问</td>\\n<td>2,960,000</td>\\n<td>724.50 亿</td>\\n<td>1,570,000.00</td>\\n<td>0.022</td>\\n<td>3.2%</td>\\n</tr><tr>\\n<td>合计</td>\\n<td>40,980,000</td>\\n<td>12,261.20<br>亿</td>\\n<td>48,688,000.00</td>\\n<td>-</td>\\n<td>100%</td>\\n</tr></table>\\n\\n**三、月度趋势（2025年）**\\n\\n<table border=\\"1\\" ><tr>\\n<td>月份</td>\\n<td>调用次数</td>\\n<td>Token 总量</td>\\n<td>费用（¥）</td>\\n<td>环比（费用）</td>\\n</tr><tr>\\n<td>1 月</td>\\n<td>2,575,000</td>\\n<td>485.00 亿</td>\\n<td>2,722,000.00</td>\\n<td>-</td>\\n</tr><tr>\\n<td>2 月</td>\\n<td>2,472,000</td>\\n<td>464.00 亿</td>\\n<td>2,605,000.00</td>\\n<td>-4.3%<br><img src=\\"https://web-api.textin.com/ocr_image/external/2a10b2c1397dab4b.jpg\\"></td>\\n</tr><tr>\\n<td>3 月</td>\\n<td>3,038,000</td>\\n<td>572.00 亿</td>\\n<td>3,205,000.00</td>\\n<td>+23.0%</td>\\n</tr><tr>\\n<td>4 月</td>\\n<td>3,263,000</td>\\n<td>614.00 亿</td>\\n<td>3,446,000.00</td>\\n<td>+7.5%</td>\\n</tr><tr>\\n<td>5 月</td>\\n<td>3,464,000</td>\\n<td>652.00 亿</td>\\n<td>3,659,000.00</td>\\n<td>+6.2%</td>\\n</tr><tr>\\n<td>6 月</td>\\n<td>3,630,000</td>\\n<td>683.00 亿</td>\\n<td>3,839,000.00</td>\\n<td>+4.9%</td>\\n</tr><tr>\\n<td>7 月</td>\\n<td>3,538,000</td>\\n<td>665.00 亿</td>\\n<td>3,733,000.00</td>\\n<td>-2.8%</td>\\n</tr><tr>\\n<td>8 月</td>\\n<td>3,721,000</td>\\n<td>700.00 亿</td>\\n<td>3,936,000.00</td>\\n<td>+5.4%</td>\\n</tr><tr>\\n<td>9 月</td>\\n<td>3,646,000</td>\\n<td>685.00 亿</td>\\n<td>3,850,000.00</td>\\n<td>-2.2%</td>\\n</tr><tr>\\n<td>10 月</td>\\n<td>3,839,000</td>\\n<td>721.00 亿</td>\\n<td>4,054,000.00</td>\\n<td>+5.3%</td>\\n</tr><tr>\\n<td>11 月</td>\\n<td>3,928,000</td>\\n<td>739.00 亿</td>\\n<td>4,153,000.00</td>\\n<td>+2.4%</td>\\n</tr><tr>\\n<td>12 月</td>\\n<td>4,146,000</td>\\n<td>780.00 亿</td>\\n<td>4,388,000.00</td>\\n<td>+5.7%</td>\\n</tr><tr>\\n<td>累计</td>\\n<td>42,260,000</td>\\n<td>7,760.00 亿</td>\\n<td>43,590,000.00</td>\\n<td>-</td>\\n</tr></table>\\n\\n<!– 440 385 月度费用 366 384 373 394 345 321 400 272 261 (A) 350 旺邮 300 250 1月 2月 3月 4月 5月 6月 7月 8月 9月 10月 11月 12月 –>\\n\\n\\n**四、Top 10用量用户排行**\\n\\n<table border=\\"1\\" ><tr>\\n<td>排名</td>\\n<td>姓名</td>\\n<td>所属部门／二级部门</td>\\n<td>API 调用次数</td>\\n<td>Token 总量</td>\\n<td>费用（¥）</td>\\n<td>人均费用排名</td>\\n</tr><tr>\\n<td>1<br><img src=\\"https://web-api.textin.com/ocr_image/external/ba7f3d967ae7e1e0.jpg\\"></td>\\n<td>钱伟豪</td>\\n<td>数据科学部／算法研究组</td>\\n<td>2,450,000</td>\\n<td>485.00 亿</td>\\n<td>1,890,000.00</td>\\n<td>1</td>\\n</tr><tr>\\n<td>2<br><img src=\\"https://web-api.textin.com/ocr_image/external/53bc114224d67844.jpg\\"></td>\\n<td>郑子涵</td>\\n<td>运营部／用户运营组</td>\\n<td>2,280,000</td>\\n<td>420.00 亿</td>\\n<td>1,750,000.00</td>\\n<td>2</td>\\n</tr><tr>\\n<td>3<br><img src=\\"https://web-api.textin.com/ocr_image/external/43d1eafd23a09124.jpg\\"></td>\\n<td>张明远</td>\\n<td>研发部／AI 平台组</td>\\n<td>2,150,000</td>\\n<td>395.00 亿</td>\\n<td>1,620,000.00</td>\\n<td>3</td>\\n</tr><tr>\\n<td>4<br><img src=\\"https://web-api.textin.com/ocr_image/external/e18b5369f3912613.jpg\\"></td>\\n<td>陆雨萱</td>\\n<td>运营部／数据运营组</td>\\n<td>1,980,000</td>\\n<td>370.00 亿</td>\\n<td>1,480,000.00</td>\\n<td>4</td>\\n</tr><tr>\\n<td>5<br><img src=\\"https://web-api.textin.com/ocr_image/external/48f179722aeb7feb.jpg\\"></td>\\n<td>周思远</td>\\n<td>市场部／内容营销组</td>\\n<td>1,850,000</td>\\n<td>340.00 亿</td>\\n<td>1,420,000.00</td>\\n<td>5</td>\\n</tr><tr>\\n<td>6<br><img src=\\"https://web-api.textin.com/ocr_image/external/ea746cadc19fece9.jpg\\"></td>\\n<td>赵雨婷</td>\\n<td>产品部／智能产品线</td>\\n<td>1,720,000</td>\\n<td>315.00 亿</td>\\n<td>1,280,000.00</td>\\n<td>6</td>\\n</tr><tr>\\n<td>7<br><img src=\\"https://web-api.textin.com/ocr_image/external/b791b3f37b98a5d8.jpg\\"></td>\\n<td>孙浩然</td>\\n<td>产品部／数据产品线</td>\\n<td>1,580,000</td>\\n<td>290.00 亿</td>\\n<td>1,180,000.00</td>\\n<td>7</td>\\n</tr><tr>\\n<td>8<br><img src=\\"https://web-api.textin.com/ocr_image/external/0c046b802aee1247.jpg\\"></td>\\n<td>李雪琴</td>\\n<td>研发部／基础架构组</td>\\n<td>1,350,000</td>\\n<td>250.00 亿</td>\\n<td>1,050,000.00</td>\\n<td>8</td>\\n</tr><tr>\\n<td>9<br><img src=\\"https://web-api.textin.com/ocr_image/external/0aa4be2f11a1a87a.jpg\\"></td>\\n<td>吴佳琪</td>\\n<td>市场部／品牌设计组</td>\\n<td>980,000</td>\\n<td>185.00 亿</td>\\n<td>780,000.00</td>\\n<td>9</td>\\n</tr><tr>\\n<td>10<br><img src=\\"https://web-api.textin.com/ocr_image/external/ed61a7dd2dacb18a.jpg\\"></td>\\n<td>王建国</td>\\n<td>研发部／测试效能组</td>\\n<td>850,000</td>\\n<td>160.00 亿</td>\\n<td>680,000.00</td>\\n<td>11</td>\\n</tr></table>\\n\\n注：王建国人均费用排名第11位，因其组内人数较少（18人），人均费用略高于部分排名靠前的用户。\\n\\n**五、项目级用量 Top 10**\\n\\n<table border=\\"1\\" ><tr>\\n<td>排名</td>\\n<td>项目名称</td>\\n<td>所属部门</td>\\n<td>负责人</td>\\n<td>API 调用</td>\\n<td>Token 总量</td>\\n<td>费用（¥）</td>\\n<td>主要使用模型</td>\\n</tr><tr>\\n<td><img src=\\"https://web-api.textin.com/ocr_image/external/06956064d4cf02d3.jpg\\"></td>\\n<td>智能客服</td>\\n<td>研发部／AI 平台组</td>\\n<td>张明远</td>\\n<td>5,800,000</td>\\n<td>892.00 亿</td>\\n<td>3,780,000.00</td>\\n<td>GPT-4/ Claude Opus</td>\\n</tr><tr>\\n<td>2<br><img src=\\"https://web-api.textin.com/ocr_image/external/f3c6386ca77d6467.jpg\\"></td>\\n<td>大模型微调</td>\\n<td>数据科学部／算法研究组</td>\\n<td>钱伟豪</td>\\n<td>5,200,000</td>\\n<td>785.00 亿</td>\\n<td>3,450,000.00</td>\\n<td>Claude Opus/GPT-4</td>\\n</tr><tr>\\n<td>3<br><img src=\\"https://web-api.textin.com/ocr_image/external/cb6fbea2a2a9ede1.jpg\\"></td>\\n<td>用户画像</td>\\n<td>运营部／用户运营组</td>\\n<td>郑子涵</td>\\n<td>4,800,000</td>\\n<td>720.00 亿</td>\\n<td>3,120,000.00</td>\\n<td>GPT-4 / Claude Sonnet</td>\\n</tr><tr>\\n<td>4<br><img src=\\"https://web-api.textin.com/ocr_image/external/dc31a7ad512f72b6.jpg\\"></td>\\n<td>AI 搜索</td>\\n<td>产品部／智能产品线</td>\\n<td>赵雨婷</td>\\n<td>4,200,000</td>\\n<td>620.00 亿</td>\\n<td>2,680,000.00</td>\\n<td>GPT -4 ／通义千问</td>\\n</tr><tr>\\n<td>5<br><img src=\\"https://web-api.textin.com/ocr_image/external/0f1482ed0eead7af.jpg\\"></td>\\n<td>内容生成</td>\\n<td>市场部／内容营销组</td>\\n<td>周思远</td>\\n<td>4,000,000</td>\\n<td>585.00 亿</td>\\n<td>2,540,000.00</td>\\n<td>GPT-4/Claude Sonnet</td>\\n</tr><tr>\\n<td>6</td>\\n<td>精准推送</td>\\n<td>运营部／用户运营组</td>\\n<td>郑子涵</td>\\n<td>3,600,000</td>\\n<td>530.00 亿</td>\\n<td>2,300,000.00</td>\\n<td>Claude Opus / GPT-4</td>\\n</tr><tr>\\n<td>7</td>\\n<td>文档理解</td>\\n<td>数据科学部／算法研究组</td>\\n<td>钱伟豪</td>\\n<td>3,400,000</td>\\n<td>500.00 亿</td>\\n<td>2,180,000.00</td>\\n<td>GPT-4/Claude Sonnet</td>\\n</tr><tr>\\n<td>8<br><img src=\\"https://web-api.textin.com/ocr_image/external/f2c2ba459ea83fed.jpg\\"></td>\\n<td>代码助手</td>\\n<td>研发部／AI 平台组</td>\\n<td>张明远</td>\\n<td>3,200,000</td>\\n<td>475.00 亿</td>\\n<td>2,050,000.00</td>\\n<td>Claude Opus/GPT-4</td>\\n</tr><tr>\\n<td>9<br><img src=\\"https://web-api.textin.com/ocr_image/external/b777c77519a0fb55.jpg\\"></td>\\n<td>A /B 测试平台</td>\\n<td>运营部／数据运营组</td>\\n<td>陆雨萱</td>\\n<td>2,900,000</td>\\n<td>430.00 亿</td>\\n<td>1,860,000.00</td>\\n<td>GPT -4 ／通义千问</td>\\n</tr><tr>\\n<td>10<br><img src=\\"https://web-api.textin.com/ocr_image/external/bd6fc7e64cf682f4.jpg\\"></td>\\n<td>智能问答</td>\\n<td>产品部／智能产品线</td>\\n<td>赵雨婷</td>\\n<td>2,700,000</td>\\n<td>400.00 亿</td>\\n<td>1,730,000.00</td>\\n<td>GPT -4 ／文心一言4 . 0</td>\\n</tr></table>\\n\\n# 六、异常告警与优化建议\\n\\n\\n\\n\\n## 内容营销组12月API调用激增\\n\\n12月份调用720,000次，较11月（680,000次）增长5.9%，较9月（630,000次）增长14.3％。建议确认是否存在非授权批量调用。\\n\\n\\n\\n\\n## 算法研究组 Claude Opus 费用严重超预算\\n\\nClaude Opus全年费用612万元，占部门总费用的41.3%，较年初预算超支12％。建议评估是否可将部分任务迁移至Claude Sonnet以降低费用。按干Token 差价0.01元计算，若迁移30％的Opus调用量，预计年节省费用约110万元。\\n\\n\\n\\n\\n## 国产模型占比偏低，成本优化空间显著\\n\\n文心一言4.0千Token均价0.026元，通义千问0.022元，均显著低于GPT-4(0.045元）和Claude Opus(0.044元）。当前国产模型调用次数占比仅18.3%，费用占比仅9.9％。用户运营组文心一言调用占比14.6%，建议全集团层面增加国产模型使用比例以控制成本。\\n\\n<!– ☒ –>\\n\\n\\n### 数据工程组Q4调用量平稳回落\\n\\nQ4月度调用量稳定在12-13万次，较Q2巅峰（11.5万次）有所回落，H2整体趋势平稳，属正常波动，无需特殊关注。\\n\\n\\n\\n\\n### 品牌设计组人均费用全集团最高\\n\\n品牌设计组15人，人均费用68,200元／年，为全公司最高（全公司均值29,000元／年，为2.35倍）。建议审查是否存在个人账号混用或多项目共用同-API Key的情况。\\n\\n制表日期：2026-01-10｜报表版本：v2.3｜下次报表生成日期：2026-02-05\\n\\n原始数据来源于集团AI网关日志系统｜本报表仅供内部管理使用\\n\\n","metadata":{"data_source":{"record_locator":{"protocol":"file","remote_file_path":"temp/某某公司AI用量报表.png"},"url":"file://temp/某某公司AI用量报表.png"},"filename":"某某公司AI用量报表.png","filetype":"image/png","page_count":1},"pages":[{"angle":0,"element_ids":["b0b28b7598462b1dc35628df9bd3086dbdfa0c4f15fea70bb9a4e328f07e137a","005d44743d4991491f5d4bada074ab5f22e4466730fc16d1bf18d30f8551f7d3","f7fe5462db30f641fafe7ab4f6c144c9310937325cb997d94fc507ca257ec08e","913ee7e62fc92de95e4d7c80b9a84ded6ef77ceb22c86649a7f556f8cad5dd45","74c3526f3d69b8129067aaf8df191f4fdba8d7e9bab184723712ce93b7ed92a1","151a7f0d9eecc434cbe14a1b79375060f837ec40226c689e12e885d3692ebeba","3718ef78786ac3fe0afc9a7ea16d69c0688a0739cfa32f5448e8f0dfa32e295b","fa37bc71e569e9fa6e29d76627af96d280c2b13bfb548fdd20af7a1765e919db","27293311abf02020ae64ebec82104dfd5d9b55a71a81cf195a3e3f88a06fe83d","6be7c91a4d9f873327966ec3f0f878de7f6002c111cf116bc7ad423f5956cb71","6ab79c8789460b2b36aabcec10b91e869d18270409a3e5c1835e95858e4f4ebd","ee9df3e9367c280e8e7f8799dc3d4a2572d455c8b11e8a945cce043f371e4702","59199dc85146506285dfc36b28b10194ccae0ddf42bf615ef100260b6cbe62bf","0d361f4d825cf1d411087be19d366b2a2cd5962a23ca9e85c772196ff835238b","4fc3d5bfc0b7071ebe9802e674ced047934dcbbf4f5ba71163cc5d76e53355f4","f85a431352d4bbaee1f2703ebb3ce1facd38e95ddbdae088cc2588bb1196c982","d890fc45661c5a8c08ea3ac8ee007eeb8842d1371b47f005031cb736130c80f4","0ed120e82f9d23db44283751a04bfe7f8d9ee7ae4d2bb97086eac6a1f2f72028","6841538746fc9f3da46a66c24cd07a07c5c0e6558a4075bcd0ce148a98b0f173","40507e4a4de837fac52069b3cd8052b958d0521b70dc11db8a4fe7b2fea5758f","45b946ff85c83be9966cc4ba388bfce6f3c4ca9351ca1b1521cc9507f915864c","d8dbc7a9e5eb026f2dffa9644d6f105a2dfee671e900b3a7cb4523244e0f7020","41096bd2e917c5105652fad5b6d174762ca9d214f08f430b9857405102eb53bd","cbf50cca607f8489d297a89badd62f1c00bdc880c7728fb13cb80882bb4cb285","8849f6dbfbe793665ebdb3e6e307d1d8797c264b9ebe06a404316ed8c44890dd","4a6dbe6256f876cf819f1d197fa3470a763b1555bdb625aff77c81a847f29333","0281ce60bf9ca5e0b5c9dd94ca193bc5116bd989e22971d986bdd27f64f34fce","4a922b435a499e99243cb96ed4fbbac91c8904ac6cf85f7ab6e418f808e737cf","d1a7efcef9dc649c3fcda506755c4144406831abe0d5dd0f423b6d5ba4dc01dc","05ae9dc9fe33dadf39ca2c13203e369a804bd771a0a84782055e86b828cc145e"],"page_height":6175,"page_image_url":"https://web-api.textin.com/ocr_image/external/965ea182ea1adbff.jpg","page_number":1,"page_width":1870,"status":"Success"}],"schema_version":"1.3.1","success_count":1,"summary":{"duration_ms":3119}}

关键下游场景：

• RAG 问答： header_tree 让 LLM 能精确回答"项目级用量 Top 10"，而不是模糊地匹配"费用"列
• 知识库构建： footnotes 和 unit 字段保留了数据的完整上下文，避免"拿到正确数字做出错误解读"
• 数据入库： 标准 JSON Schema 可直接对接数据库写入流程，无需二次清洗

准确回答，无一字错漏 👍

六、总结

回到最初的问题：表格解析结果到手了，怎么判断能不能用？

答案是三层递进式检查：

在本文的实测中，TextIn xParse 在三个维度、三个真实业务案例中的表现均达到了可用标准，尤其是在复杂合并单元格工程表（案例1）、嵌套多个表的劳务报酬审批单（案例2）、低对比度小字识别（案例3）这三个其他方案普遍出问题的场景中，xParse 给出了正确的解析结果。

体验链接（注册送1000页体验额度）：https://www.textin.com/register/code/VT4KAZ

本文测评数据基于 2026年6月 实测结果，各开源方案版本为当时最新稳定版。产品功能以 TextIn xParse 官方最新版本为准。

在本文的实测中，TextIn xParse 在三个维度、三个真实业务案例中的表现均达到了可用标准，尤其是在复杂合并单元格工程表（案例1）、嵌套多个表的劳务报酬审批单（案例2）、低对比度小字识别（案例3）这三个其他方案普遍出问题的场景中，xParse 给出了正确的解析结果。

体验链接（注册送1000页体验额度）：https://www.textin.com/register/code/VT4KAZ

本文测评数据基于 2026年6月 实测结果，各开源方案版本为当时最新稳定版。产品功能以 TextIn xParse 官方最新版本为准。

结束语

        亲爱的朋友，无论前路如何漫长与崎岖，都请怀揣梦想的火种，因为在生活的广袤星空中，总有一颗属于你的璀璨星辰在熠熠生辉，静候你抵达。

         愿你在这纷繁世间，能时常收获微小而确定的幸福，如春日微风轻拂面庞，所有的疲惫与烦恼都能被温柔以待，内心永远充盈着安宁与慰藉。

        至此，文章已至尾声，而您的故事仍在续写，不知您对文中所叙有何独特见解？期待您在心中与我对话，开启思想的新交流。

————— 业精于勤，荒于嬉 —————

————— 行成于思，毁于随 —————

     💞 关注博主 带你实现畅游前后端

     🏰 大屏可视化 带你体验酷炫大屏

     💯 神秘个人简介 带你体验不一样得介绍

     🎀 酷炫邀请函 带你体验高大上得邀请

     亲，码字不易，动动小手，欢迎 点赞 ➕ 收藏，如 🈶 问题请留言（评论），博主看见后一定及时给您答复，💌💌💌

原文地址：https://blog.csdn.net/weixin_43151418/article/details/162234509（防止抄袭，原文地址不可删除）