Claude launches file creation for professional documents

Anthropic released file creation capabilities for Claude this week, enabling users to generate professional documents through natural language prompts. The feature became available as a preview to paid subscribers across Pro, Max, Team, and Enterprise plans on web, desktop, and mobile platforms.

The announcement, documented in Anthropic's support center with an "Updated this week" timestamp, marks a technical expansion of Claude's computing environment. Users can now request Excel spreadsheets with functional formulas, PowerPoint presentations, Word documents, and PDF files without specialized software knowledge. Claude Sonnet 4.5 powers the file creation system, which Anthropic describes as excelling at document manipulation tasks.

File generation operates through a sandboxed computing environment with limited internet access. Users describe requirements in natural language, and Claude produces downloadable files directly within conversations. The system supports files up to 30MB for both uploads and downloads. Files can be saved to Google Drive as an alternative to direct downloads.

The capability extends beyond basic document creation to data analysis workflows. Claude can process CSV and TSV files, create Python scripts for data analysis, and generate PNG visualization files. Users can upload datasets for analysis and request visualizations, with Claude handling the technical implementation autonomously.

Anthropic documented common workflows in its support materials. Financial modeling represents one application, where users can request "a monthly budget tracker with income, expenses categories, and automatic calculations for savings." The system produces Excel files with formulas, formatting, and charts. Report generation combines data analysis with document creation, enabling requests like "Create a quarterly sales report using this CSV data, including trend analysis and recommendations."

Format conversion capabilities allow users to transform documents between file types. A Word document can become a PowerPoint presentation through a simple request. More complex workflows involving multiple format conversions received explicit documentation: users might upload a CSV file and request Claude to create a financial model, write a summary memo, and generate a presentation from the results.

PDF data extraction illustrates another technical capability. Users can upload PDF files containing tables or forms and request extraction into Excel spreadsheets with summary charts. Claude processes the document structure, organizes data in spreadsheet format, and adds visualizations.

Machine learning applications enter documented use cases. Users can upload CSV files with datasets and request Claude to build predictive models for specific outcomes. The system uses Python to train models and provides reports explaining methodology, model quality, and results.

The feature requires manual activation. Enterprise organizations have file creation disabled by default at the organizational level. Owners must enable it through Admin settings by toggling "Code execution and file creation" on, after which individual members still need to opt into the feature through their personal Settings. Team plans have the feature enabled by default at the organizational level, but individual members must still manually activate it. Pro and Max subscribers enable file creation directly through Settings.

Mobile implementation maintains functionality across iOS and Android platforms. Users access the toggle through Settings by selecting their name or initials from the sidebar. Downloaded files open in system preview applications or dedicated apps like Microsoft Word for .docx files.

Activating file creation automatically disables Claude's previous "Analysis tool," which operated on JavaScript. The new Python-based system expands capabilities beyond the earlier tool's functionality. Artifacts—the feature that renders HTML, React applications, markdown documents, mermaid diagrams, and SVG files within the interface—continue working with file creation enabled, though the user experience changed as Claude now uses the computing environment for artifact generation.

Security considerations received detailed documentation. The sandboxed environment includes internet access for package downloads through NPM and similar services. Anthropic identified two primary risk vectors: bad actors could embed instructions in external files or websites that trick Claude into downloading and executing untrusted code, or the system could be manipulated into reading sensitive data from connected sources and making external network requests to transmit that information.

Anthropic documented specific security mitigations. Users receive full control over the feature with on-off toggle access. The system provides summaries of Claude's actions for monitoring purposes. Users can review and audit actions taken within the sandbox environment. Public sharing of conversations containing file artifacts remains disabled for Pro and Max users. Task duration and sandbox container lifetime face intentional limitations to prevent loops of malicious activity. Sandbox environments never share between users, and network, container, and storage resources have explicit restrictions. A classifier detects prompt injections and stops execution when threats are identified.

The company conducted red-teaming and security testing before release and maintains ongoing testing processes. Team and Enterprise owners receive organizational-level control, allowing them to toggle the feature on or off for their entire organization. Anthropic notes that Claude can only be manipulated into leaking data it accesses through individual user prompts, projects, or activated connections.

The system maintains an allowlist of approved network destinations. Allowed domains include api.anthropic.com, statsig.anthropic.com, github.com, registry.npmjs.org, npmjs.com, npmjs.org, yarnpkg.com, registry.yarnpkg.com, pypi.org, files.pythonhosted.org, pythonhosted.org, archive.ubuntu.com, and security.ubuntu.com. This whitelist approach limits potential data exfiltration vectors while maintaining access to necessary development resources.

Usage impacts vary from standard chat interactions. Anthropic disclosed that file creation consumes more usage quota compared to normal conversations with Claude, though the company did not specify exact ratios. Usage draws from the same limits associated with each plan tier.

Multi-file capabilities extend across single conversations. Users can work with multiple files simultaneously, enabling comprehensive reporting and analysis. Files remain available for download throughout the conversation session.

The feature arrives as browser-based AI capabilities expand across the industry. Anthropic launched Claude for Chrome as a research preview in August 2025, while competitors including Perplexity introduced standalone browsers with integrated AI functionality. Document generation represents a different approach to productivity enhancement compared to browser automation or web navigation capabilities.

Previous Claude developments established groundwork for file handling. Artifacts received expansion in June 2025 to support AI-powered application development, enabling users to build tools that embed Claude's intelligence. The artifacts system generated over half a billion creations according to Anthropic's June announcement. Integration capabilities launched in May 2025 connected Claude to workplace applications through the Model Context Protocol, enabling access to services including Jira, Confluence, Asana, and Linear.

Technical documentation emphasizes the distinction between Claude's knowledge cutoff and real-time capabilities. The file creation feature operates on uploaded data and user requests rather than accessing current information beyond its training data. Users requiring current market data or recent statistics for financial models would need to provide that information through uploads or integration connections rather than relying on Claude's base knowledge.

The marketing technology implications extend to document workflow automation. Teams generating regular reports, presentations, and analysis documents can describe requirements to Claude rather than manually constructing files. Data visualization requests that previously required specialized software or coding knowledge become accessible through natural language prompts. The system handles formatting, formula creation, and chart generation based on conversational descriptions.

Enterprise adoption considerations center on security configurations and organizational controls. The organizational-level toggle for Team and Enterprise plans allows administrators to evaluate security postures before enabling the feature broadly. Individual user activation requirements provide a secondary control layer, ensuring team members consciously choose to use capabilities that involve code execution and file manipulation.

Performance characteristics remain undefined in public documentation. Anthropic did not specify response times for file generation requests, computational resource allocation per user, or queue management during high-demand periods. File size limitations at 30MB per file establish upper boundaries for dataset processing and output generation.

Python package management received specific documentation. Installation requires the --break-system-packages flag for all pip installations, reflecting the sandboxed environment's configuration. NPM packages install to /home/claude/.npm-global directory. Virtual environments become available for complex Python projects requiring dependency isolation.

The feature's positioning as a preview indicates ongoing development. Anthropic solicits user feedback through the thumb up/down functionality within claude.ai, suggesting iterative improvements based on practical usage patterns. The company's documentation notes users should "review and refine Claude's outputs to meet your exact requirements," acknowledging that generated files may require manual adjustments.

Competition in AI-powered document generation intensified throughout 2025. Microsoft integrated similar capabilities into Copilot, while Google expanded Bard's document handling. Anthropic's approach through sandboxed computing environments with explicit security controls differentiates the implementation from competitors using different architectural approaches.

The maximum file size restriction at 30MB limits certain use cases. Large datasets, high-resolution image collections, and extensive document compilations may require splitting across multiple files or external processing. The limitation applies equally to uploads and downloads, creating symmetrical constraints on input and output operations.

Financial modeling capabilities documented in support materials suggest applications for budget planning, forecasting, and analysis. The ability to generate working formulas distinguishes the feature from simple template filling, as Claude constructs calculation logic based on described requirements. Charts and visualizations embedded in spreadsheets provide immediate data representation without requiring separate visualization tools.

Timeline

• October 22, 2024: Claude introduces computer use capability in public beta
• May 2025: Anthropic introduces Integrations feature connecting Claude to workplace applications
• June 25, 2025: Claude artifacts expand to enable AI-powered app creation
• July 15, 2025: Anthropic unveils comprehensive financial analysis platform with Claude
• July 29, 2025: HubSpot launches first CRM connector for Anthropic's Claude
• August 26, 2025: Anthropic launches Claude for Chrome extension research preview with 1,000 users
• October 2025: Anthropic releases file creation capabilities for Claude as feature preview across paid plans

Summary

Who: Anthropic released the file creation feature for Claude users with paid subscriptions including Pro, Max, Team, and Enterprise plans. The feature became available to millions of existing Claude users across web, desktop, and mobile platforms.

What: A file creation capability enabling users to generate Excel spreadsheets, PowerPoint presentations, Word documents, and PDF files through natural language prompts. The system operates through a sandboxed computing environment powered by Claude Sonnet 4.5, supporting files up to 30MB. Capabilities extend to data analysis, Python script creation, CSV/TSV file processing, and machine learning model building. Users can convert between file formats, extract data from PDFs, and generate visualizations.

When: The feature received documentation marked "Updated this week" in Anthropic's support center, indicating release in October 2025. The announcement did not specify an exact date.

Where: The capability deployed globally across Claude's web interface, Claude Desktop applications, and Claude Mobile for iOS and Android. Files can be downloaded directly or saved to Google Drive. The feature operates within a sandboxed computing environment with restricted internet access limited to approved domains including GitHub, NPM registries, Python package repositories, and Ubuntu archives.

Why: The feature addresses demand for document automation and eliminates requirements for specialized software skills in spreadsheet creation, presentation design, and data analysis. Organizations can automate routine document generation, perform complex analyses without coding knowledge, and streamline workflows involving multiple file format conversions. The sandboxed approach with security controls aims to balance functionality with risk mitigation for enterprise deployments. Usage of the feature consumes more quota than standard conversations, aligning with Anthropic's monetization of compute-intensive capabilities across paid subscription tiers.