That distinction has widened as OpenAI turns ChatGPT into a work hub around Projects, Apps, Canvas, deep research, and Codex, while Anthropic has built Claude around Projects, Artifacts, desktop extensions, connectors, and Claude Code. The overlap is real. The operational model is not.
Engadget presents the choice chiefly as a question of matching an LLM to a task. That is fair for personal use. But for anyone putting these tools on a Windows workstation, choosing an assistant by its prose style or a one-off coding test misses the control points that create the real risk: what data is attached, where the assistant can retrieve context, which actions it can take, how its desktop client is deployed, and whether staff understand the difference between a chat window and an agent with access to business systems.
ChatGPT and Claude Have Converged at the Workspace Layer
The older shorthand—ChatGPT for general-purpose use, Claude for long documents and careful writing—has become less useful. Both products now provide persistent work areas where users can collect files, set instructions, retain context, and return to a task without rebuilding the prompt every time.
OpenAI’s Projects are available across ChatGPT account tiers, according to the company’s current help documentation. A project can contain chats, uploaded material, and tailored instructions; paid accounts may add tools such as agent mode and deep research. ChatGPT also has Canvas, a side-by-side editor for writing and code that supports direct edits, revision history, inline suggestions, code review, exports, and Python execution.
Anthropic’s Projects use the same basic pattern: a dedicated workspace, its own chat history, project instructions, and a knowledge base assembled from documents, text, code, or other uploaded material. Anthropic says its project knowledge can switch to retrieval-augmented generation when the stored material approaches context limits, extending the effective amount of reference material that Claude can use.
For a user trying to build a policy draft, analyze a pile of PDFs, or maintain a scripting project, either approach is competent. The practical difference is in the center of gravity. ChatGPT treats the project as one component in an increasingly broad product that includes custom GPTs, connected apps, voice, image generation, research features, and agents. Claude’s product remains more oriented around a focused conversation, a project knowledge base, and a document or tool produced from that conversation.
That makes ChatGPT more attractive when a team wants one subscription to cover many different kinds of work. Claude can feel less cluttered for people who want an assistant to stay with a document, a codebase, or a collection of internal reference material without constantly presenting another mode to try.
There is a documentation wrinkle here that administrators should notice before assuming ChatGPT’s free tier lacks workspace capability. OpenAI’s current Projects help page says Projects are available to all free and paid account types, while the feature matrix on its pricing page still lists Projects as unavailable on Free. The help article is newer and describes how the feature currently works, but the inconsistency is a reminder that AI-plan comparison tables age faster than the products they describe. IT buyers should verify entitlements in the tenant they intend to deploy rather than relying on an overview page or a comparison article published even a few weeks earlier.
Windows Deployment Is a More Concrete Divider Than Model Personality
ChatGPT and Claude both have Windows desktop applications, but their delivery and integration paths differ in ways that matter to managed environments.
OpenAI distributes the ChatGPT Windows app through the Microsoft Store. Its own documentation says the application requires Windows 10 version 17763.0 or later on x64 or Arm64 hardware, and that organizations can deploy it with management tools such as Microsoft Intune or install it with
winget. The app includes a companion window invoked by a configurable keyboard shortcut, making it easy for a user to pull ChatGPT over whatever is on screen.
That convenience is useful, but it also makes policy decisions less theoretical. A company that blocks Microsoft Store installs needs a deployment plan. A company that permits ChatGPT needs to decide whether users may attach internal files, enable connected Apps, use browser features, or carry material from a corporate ChatGPT workspace into a personal account. The application itself is not the governance boundary; identity, workspace controls, data classification, and endpoint policy are.
Anthropic’s Claude Desktop is currently labeled beta and supports Windows 10 or later. More important, it has desktop extensions designed to connect Claude with local applications, files, calendars, email, messaging tools, and other system resources. Anthropic distinguishes these local extensions from remote web connectors, which communicate with cloud services.
This is where the two products stop being interchangeable. Claude Desktop’s extension model may be particularly appealing to power users who want the assistant close to local workflows. But giving an assistant access to a filesystem, mail client, or a locally available tool is a larger security decision than uploading a single document into a chat. Administrators need an approved-extension list, a process for reviewing what permissions an extension requests, and a clear rule for whether personal Claude accounts may connect to corporate resources.
Neither product’s desktop client should be deployed as if it were a passive note-taking application. They are increasingly gateways between a user’s prompt and data that may live in SharePoint, Google Drive, GitHub, a local folder, a business app, or an internal service.
Connected Data Is Where the Choice Gets Expensive
ChatGPT has renamed its connectors as “Apps,” consolidating tools that can search connected information, synchronize data into a workspace, present interactive interfaces, or in some cases take actions. OpenAI says Apps can be available to logged-in users, but individual app capabilities and availability depend on plan and region. Business and Enterprise workspaces add administrative controls around this environment.
Claude offers remote connectors as well as local desktop extensions, including support for the Model Context Protocol approach to connecting tools and data sources. Paid Claude plans can connect services such as Google Workspace, while Anthropic’s higher-end business offerings extend the management and security story.
The visible product comparison can make this look like feature parity: both assistants can “connect your tools.” The implementation question is more important than the marketing phrase. A connector that only searches a repository has a different threat model from one that can write data, create tickets, send messages, modify records, or trigger automation. A synced repository of internal documents is different again, because it may make information available long after the user who enabled the integration has forgotten which files were indexed.
For a Windows-focused organization, the right first deployment is usually narrower than vendors encourage. Start with a sanctioned business workspace, a limited set of read-only sources, a small user group, and test data that resembles real work without being regulated or highly sensitive. Log what the product accessed, what users exported, and which capabilities they actually used. Then decide whether the productivity gain justifies adding write-capable actions or local extensions.
A user who wants an assistant for personal research can reasonably choose based on interface preference. A company that is about to connect SharePoint, GitHub, Outlook data, or local work folders needs a procurement and security review.
Coding Claims Need Tests, Not Brand Loyalty
Both companies now sell more than conversational code help. Anthropic positions Claude Code as a tool that works directly from the terminal, while OpenAI packages Codex alongside its broader ChatGPT work environment. Independent coding comparisons, including recent testing published by Zapier and Tom’s Guide, have found meaningful differences in workflow and task outcomes, but they do not establish a permanent winner.
That is not evasiveness; it reflects the reality of the products. Results change with the underlying model, the agent version, tool permissions, repository structure, rate limits, test harness, and whether an assistant can actually run code and inspect the result. A coding model that produces a convincing patch in a browser can fail badly when asked to understand a Windows-specific build chain, PowerShell module, legacy .NET application, Group Policy dependency, or signing process.
The useful evaluation is therefore local. Give Claude Code and Codex the same bounded job: repair a known defect in a disposable repository, add tests, run the tests, explain the change, and submit a patch that a senior developer can review. Measure setup friction, correctness, test coverage, rollback quality, secret handling, and how often the agent changes unrelated files. Do not measure only whether its explanation sounds confident.
For Windows administrators, the same principle applies to scripting. Test PowerShell generation in a non-production tenant and require the assistant to produce a dry-run path, error handling, logging, and explicit scope controls. A chatbot that generates a one-line command to remove software at scale is not delivering an automation solution; it is creating a change-management problem.
Privacy Controls Are Configuration, Not a Product Badge
Consumer ChatGPT users can disable “Improve the model for everyone” in Data Controls so new chats are not used to train OpenAI’s models. OpenAI also offers Temporary Chat, which does not appear in history, does not create memories, and is not used for model improvement, though the company says those conversations are retained for 30 days for safety purposes.
OpenAI says content in its API, ChatGPT Business, and ChatGPT Enterprise offerings is not used to improve model performance by default. That makes a business workspace materially different from an employee’s personal ChatGPT account, even if the two interfaces look almost identical.
Anthropic similarly separates consumer and business contexts in its product and privacy materials, but customers should treat Claude’s Projects, connectors, and desktop extensions as data-handling choices that require the same review. A project knowledge base is useful precisely because it preserves organizational context. The more useful it becomes, the more important it is to know who can access it, how long it persists, whether it can be shared, and what happens when an employee leaves.
The decision therefore is not “Claude is private” or “ChatGPT is private.” Both vendors offer controls; both expose more data when users enable more context and more integrations. The defensible choice is the plan, tenant configuration, identity model, retention terms, and extension policy that match the organization’s data.
ChatGPT remains the broader default for users who want a single AI service spanning research, multimedia work, connected apps, custom assistants, and agent-style tasks. Claude is a strong fit for users who value its project-centered interface, Artifacts, and desktop-extension path. But the feature race has made the old binary less important. The winning deployment will be the one that treats AI access as an endpoint, identity, and data-governance decision before it treats it as a chatbot preference.
References
- Primary source: Engadget
Published: August 8, 2026 at 3:30 PM UTC
Loading…
www.engadget.com - Related coverage: openai.com
Loading…
openai.com - Related coverage: help.openai.com
Loading…
help.openai.com - Related coverage: help.openai.com
Loading…
help.openai.com - Related coverage: anthropic.com
Plans & Pricing | Claude by Anthropic
Choose the Claude plan that fits how you solve problems. Free, Pro, Max, Team, and Enterprise tiers, plus API pricing for developers.www.anthropic.com - Related coverage: openai.com
ChatGPT is now a partner for your most ambitious work | OpenAI
ChatGPT Work is an agent that can take action across your apps and files, stay with a project for hours if needed, and turn a goal into finished work.openai.com - Related coverage: cdn.openai.com
- Related coverage: resources.anthropic.com
- Related coverage: go.openai.com
- Related coverage: zapier.com
Loading…
zapier.com - Related coverage: tomsguide.com
Loading…
www.tomsguide.com - Related coverage: dupple.com
Loading…
dupple.com - Related coverage: automationatlas.io
Loading…
automationatlas.io - Related coverage: emergent.sh
Loading…
emergent.sh - Related coverage: aieatingtheworld.com
Loading…
www.aieatingtheworld.com - Related coverage: devaireviews.com
Loading…
devaireviews.com - Related coverage: xtremedev.net
Loading…
xtremedev.net - Related coverage: tomsguide.com
Loading…
www.tomsguide.com - Related coverage: techradar.com
Loading…
www.techradar.com - Related coverage: theatlantic.com
Loading…
www.theatlantic.com - Related coverage: time.com
AI Is Moving Past Chatbots. Claude Cowork Shows What's Next
An AI powerful enough to analyze DNA, file taxes, and grow tomato plants is being redesigned for everyday work, pointing toward life beyond chatbots.time.com - Related coverage: itpro.com
Anthropic's Claude Opus 4.5 model stakes its claim as the the new frontrunner for AI coding | IT Pro
Anthropic has announced Claude Opus 4.5, its most advanced model to date and the new industry leader for AI code generation.www.itpro.com - Related coverage: support.anthropic.com
Install Claude Desktop | Claude Help Center
support.anthropic.com