Uni-Parser 1.5：从整篇文档到单元素解析，让科学数据更易获取
从一篇论文中提取表格，从专利中识别分子结构，或把一张反应路线截图交给科研 Agent——科学数据的获取，既需要完整的文档解析，也需要针对局部内容的灵活处理。
在 Uni-Parser 1.4 中，我们围绕复杂科学版面、多模态内容和化学信息提取，完善了从 PDF 到结构化数据的处理能力。Uni-Parser 1.5 进一步把这套能力扩展到单个内容元素，同时正式发布 MolParser 2.0 GA（General Availability，正式可用版），让复杂化学结构的识别、表达与后续使用更加完整。
本次更新围绕三个方向展开：
- 更灵活地获取数据：新增文字、表格、分子结构、图表、化学反应式和数学公式的独立解析，支持文字、表格与反应式中的混合内容识别。
- 更准确地提取化学信息：MolParser 2.0 GA 采用 E-SMILES 2.0 规范，增强立体化学、Markush 与高分子支持，完善缩写代入和分子绘图。
- 更方便地使用解析结果：新增文件分享入口与图片托管输出，接入玻尔钱包支付，完善余额展示与用量查询。
从整篇文档到一张截图，让解析贴近实际工作
完整 PDF 适合构建论文库和专利库，但在日常科研中，待处理的内容也可能只是一张表格、一幅分子结构图，或一段包含公式的文字截图。为了获取其中的信息，往往还需要裁图、分开识别，再把结果重新整理到一起。
Uni-Parser 1.5 新增单元素解析，支持独立提交文本、表格、分子结构、图表、化学反应式和数学公式图片。 首页已提供标有“内测中”的单元素解析入口：选择“单元素解析”模式，上传图片后，在上传区右下角确认类型与质量，再点击“开始解析”。默认使用高质量，也可切换为快速；上传图片不会自动提交解析。
开发者也可通过 POST /trigger-item-async HTTP API 指定类型与质量，将局部图片接入数据处理流程；参数、后端要求与完整 Python 示例见 单元素解析使用与 API 接入说明。
更重要的是，科学文档中的元素经常混合出现：文字里有公式，表格里有分子，反应式中同时包含多个化学结构。本次更新支持识别文字与表格中的分子和公式，以及化学反应式中的分子结构，减少手动拆分与多次提交。首页会根据所选类型自动启用内部元素识别，无需额外设置。
这让同一套解析能力可以服务于不同规模的任务：处理整篇文献时保留全文结构，处理局部图片时聚焦需要的内容。例如，从专利中截取一张含分子结构的表格，即可继续提取表格内容及其中的化学信息，用于后续整理、检索与分析。
MolParser 2.0 正式发布：让复杂化学结构进入数据链路
MolParser 2.0 GA 随 Uni-Parser 1.5.0 正式发布，采用 E-SMILES 2.0 规范，更新连接点与高分子支持，并扩展特殊 Markush、多环不定连接、子结构复制、虚拟连接弧和轴手性的表达。
正式版在 BioVista OCSR Benchmark 上的准确率由 preview 的 84.5% 提升至 85.7%，并改善复杂立体结构识别、缩写与 Markush 代入、置信度和分子绘图。配套的 MolParser Utils 与 Skills 为后处理、格式转换、基团代入、绘图与校验提供开源工具。
完整的语法、58 张原图与对照、工具调用示例，以及九类结构识别对比，已整理为站内专题：MolParser 2.0：端到端分子识别升级。
让解析结果更易分享、更易接入
当解析结果需要交给同事检查，或被另一个系统引用时，便捷的共享方式同样重要。Uni-Parser 1.5 在结果页面新增了整份文件的分享链接入口，可以一键复制链接，方便协作查看。
对于结果中的图片，本次新增图片托管输出：支持上传至对象存储并返回访问链接，便于下游系统引用；上传失败时自动回退为 Base64 输出，保留图片内容。
与此同时，我们改善了文本间距、中英文混排空格与 Markdown 多余转义字符，修复空表格或异常表格导致的输出失败、表格内公式显示与原有空行误删等问题。页面定位和局部重解析也进一步完善，减少分子图像重叠、尺寸不协调、局部内容遗漏与页码对应错误，让结果阅读与检查更加顺畅。
对于持续运行的任务，本次还改善了异步任务状态和结果查询的兼容性、远程文件下载重试，以及批量分子识别在超时或工作进程异常后的恢复能力。
玻尔钱包支付与用量查询：支持持续的数据处理
从少量试用到批量任务，用户需要清楚地了解可用余额和任务消耗。Uni-Parser 1.5 在已开通钱包支付的环境中，支持使用玻尔钱包支付解析费用，并完善试用额度、钱包余额及用量信息的展示与查询。
未指定付款方式时，优先使用足以支付整笔费用的 UniParser 余额；本地余额不足且满足钱包使用条件时，使用玻尔钱包。文档解析按每页 0.05 试用额度计费，无其他额外计费；新增单元素解析按每次 0.01 试用额度计费。
账户查询 API 支持按日期、付款来源等条件查询用量、付款记录和余额流水，并按日汇总请求数、解析页数与费用，方便跟踪批量任务的消耗。本次也修复了并发或重试场景下的重复扣款风险、支付后任务未继续处理，以及部分余额判断和跨时区统计问题，完善任务结果与账单的访问权限校验。
开始使用 Uni-Parser 1.5
在线体验与玻尔登录
访问 Uni-Parser 在线 Demo，使用玻尔账号登录并授权，或输入已有 API Key，即可进入解析工作流。登录后可以查看试用额度与可用的 Bohrium 钱包余额。
使用玻尔账号登录 Uni-Parser、查看余额与充值的图文步骤已整理为站内玻尔用户使用指南：登录、钱包余额与充值，可随时查阅。
CLI 与开发者接入
对于脚本、批处理、notebook 和 Agent 应用，可以通过 API 文档或开源工具箱 UniParser-Tools接入。已有的文档解析工作流可以继续使用 CLI：
pip install "git+https://github.com/dptech-corp/UniParser-Tools.git"
uniparser auth
uniparser parse paper.pdf
任务中断或需要重新获取结果时，可使用保存的 token：
uniparser fetch --token "TASK_TOKEN"
UniParser-Tools 也提供 Agent Skill，可将文档解析接入支持 Skill 的编程助手：
npx skills add dptech-corp/UniParser-Tools
从一篇完整文献到一张局部截图，从分子结构识别到后续计算，Uni-Parser 1.5 继续让科学资料中的信息成为可获取、可检查、可复用的数据。
- 在线体验：Uni-Parser
- API 文档：开发者文档
- 文档解析工具：UniParser-Tools
- 分子处理工具：MolParser
- 私有化部署与商业咨询：uniparser@dp.tech
Uni-Parser 1.5: From Full Documents to Individual Elements
Extracting a table from a paper, recognizing a molecular structure in a patent, or passing a reaction-scheme screenshot to a research Agent all require access to scientific data. That work calls for both complete document parsing and flexible processing of individual pieces of content.
Uni-Parser 1.4 improved the path from PDFs to structured data through better scientific-layout understanding, multimodal parsing, and chemical-information extraction. Uni-Parser 1.5 extends those capabilities to individual elements and officially introduces MolParser 2.0 GA (General Availability), improving how complex chemical structures are recognized, represented, and used downstream.
This release focuses on three areas:
- More flexible data extraction: Independent parsing of text, tables, molecular structures, charts, reaction schemes, and mathematical equations, with recognition of mixed content within text, tables, and reaction schemes.
- More accurate chemical-information extraction: MolParser 2.0 GA adopts E-SMILES 2.0, strengthening stereochemistry, Markush, and polymer support alongside abbreviation substitution and molecular rendering.
- Easier use of parsing results: File-sharing links, hosted image output, Bohrium wallet payments, and improved balance display and usage queries.
From a Full Document to a Single Screenshot
Complete PDFs are useful for building paper and patent collections. In everyday research, however, the content to process may be just one table, a molecular drawing, or a text screenshot containing equations. Extracting that information can require cropping, separate recognition steps, and manual assembly of the results.
Uni-Parser 1.5 adds individual-element parsing for text, tables, molecular structures, charts, chemical reaction schemes, and mathematical equation images. The home page now offers an individual-element entry marked “Beta”: select “Individual Element”, upload an image, confirm its type and quality at the bottom right of the upload area, then click “Start parsing”. High Quality is selected by default, with Fast also available. Uploading does not automatically submit a parsing task.
Developers can also specify type and quality through the POST /trigger-item-async HTTP API to integrate individual images into data workflows. See the individual-element usage and API guide for parameters, backend requirements, and a complete Python example.
Scientific elements also frequently appear together: equations within text, molecules within tables, and multiple chemical structures within a reaction scheme. This release supports recognition of molecules and equations within text and tables, as well as molecular structures within reaction schemes, reducing manual splitting and repeated submissions. The home page automatically enables nested-element detection for these types, without extra settings.
The same parsing capabilities can therefore support different task sizes: preserving full-document structure for complete literature processing, or focusing on selected content in a cropped image. A table cropped from a patent, for example, can be submitted to extract both its table content and the chemical information it contains for further organization, retrieval, and analysis.
MolParser 2.0 GA: Bringing Complex Chemistry into the Data Pipeline
MolParser 2.0 GA is officially released with Uni-Parser 1.5.0, adopting E-SMILES 2.0 with updated attachment-point and polymer support and expanded representation of special Markush structures, multi-ring uncertainty, substructure replication, virtual arcs, and axial chirality.
Accuracy on the BioVista OCSR Benchmark improves from 84.5% in preview to 85.7%. The release also improves complex stereochemistry recognition, abbreviation and Markush substitution, confidence scoring, and rendering. MolParser Utils and accompanying Skills provide open-source tools for post-processing, format conversion, group substitution, rendering, and validation.
The complete syntax, 58 original images and comparisons, tool examples, and recognition comparisons across nine structure types are available in the MolParser 2.0 and E-SMILES 2.0 Guide on this site.
Results That Are Easier to Share and Integrate
When a colleague needs to review a result or another system needs to reference it, convenient sharing matters. Uni-Parser 1.5 adds an entry point on the results page to copy a sharing link for the entire file.
This release also adds hosted image output. Result images can be uploaded to object storage and returned as accessible links for downstream use. If an upload fails, output automatically falls back to Base64, preserving the image content.
We have also improved word spacing, spacing in mixed Chinese and English text, and unnecessary Markdown escapes. Fixes address formatting failures caused by empty or malformed tables, equation display within tables, and accidental removal of original blank rows. Better positioning and partial re-parsing reduce overlapping or poorly sized molecular images, missing content, and page-mapping errors, making results easier to read and review.
For ongoing workloads, the release improves asynchronous task-status and result-query compatibility, remote-file download retries, and recovery of batch molecular recognition after timeouts or worker failures.
Bohrium Wallet Payments and Usage Queries
Moving from initial trials to batch workloads requires a clear view of available balances and task consumption. In environments where wallet payments are enabled, Uni-Parser 1.5 supports paying parsing fees through the Bohrium wallet, with improved display and queries for trial credits, wallet balances, and usage.
When no payment method is specified, Uni-Parser first uses its local balance if that balance covers the entire charge. If the local balance is insufficient and wallet eligibility conditions are met, it uses the Bohrium wallet. Document parsing costs 0.05 trial credit per page, with no additional charges; individual-element parsing costs 0.01 trial credit per request.
Account APIs support queries for usage, payment records, and balance transactions filtered by date and payment source, plus daily summaries of requests, parsed pages, and costs. This release also fixes duplicate-charge risks during concurrent requests or retries, tasks failing to resume after payment, and some balance-check and cross-time-zone aggregation issues, while strengthening access checks for task results and bills.
Get Started with Uni-Parser 1.5
Online Experience and Bohrium Sign-In
Visit the Uni-Parser online Demo, sign in with a Bohrium account and authorize access, or enter an existing API Key to start your parsing workflow. Trial credits and the available Bohrium wallet balance are shown after sign-in.
For illustrated steps on using Uni-Parser with a Bohrium account, including sign-in, balance checks, and top-ups, see the Uni-Parser Guide for Bohrium Users on this site.
CLI and Developer Integration
For scripts, batch processing, notebooks, and Agent applications, use the API documentation or the open-source UniParser-Tools toolkit. Existing document parsing workflows can continue to use the CLI:
pip install "git+https://github.com/dptech-corp/UniParser-Tools.git"
uniparser auth
uniparser parse paper.pdf
If a task is interrupted or its results need to be fetched again, use the saved token:
uniparser fetch --token "TASK_TOKEN"
UniParser-Tools also provides an Agent Skill for coding assistants that support Skills:
npx skills add dptech-corp/UniParser-Tools
From a complete paper to a single screenshot, and from molecular recognition to subsequent computation, Uni-Parser 1.5 continues to make scientific information easier to extract, inspect, and reuse as data.
- Online experience: Uni-Parser
- API documentation: Developer documentation
- Document parsing tools: UniParser-Tools
- Molecular processing tools: MolParser
- Private deployment and business inquiries: uniparser@dp.tech