ChatGPT Chat is the better default when one answer completes the task. ChatGPT Work is better suited when multiple sources, files, and review steps must stay connected until a reviewable deliverable is ready. In our matched hands-on tests, both modes completed a short writing task on the first response, but Work made the research, document-building, and verification stages easier to follow in a three-source report.
If you are deciding whether Work is worth the additional time and usage it may require, this guide offers a practical decision rule rather than another feature list. We tested Chat and Work with the same account, model, prompts, source files, output language, and completion criteria as closely as the two interfaces allowed, then compared their workflows, editable files, processing time, and visible usage.
You will learn when Chat is enough, when Work earns its place, and how to choose between them for a real task. The article also covers access on the web, desktop, iPhone, and Android, along with the plan, model, and usage checks to make before relying on Work.
As one additional observation, weekly usage remaining changed from 96% to 82% during one mixed Work session lasting about 1 hour 7 minutes. That result shows a measured change on one account — not a fixed consumption rate for Work or any individual task.
This guide is for you if you:
- Need to decide whether a real task should start in Chat or Work.
- Work with multiple sources or need an editable document, spreadsheet, presentation, or report.
- Want to check Work availability, device support, and usage limits before depending on it.
- Each matched task was run once per mode; this is not an average-performance benchmark.
- Memory was enabled and may have influenced both outputs.
- Human fact-checking and revision time were not measured.
- The separate usage observation covered one mixed session, not either matched task in isolation.
- Model availability, limits, and interface labels may change.
- What Is ChatGPT Work? Key Differences from Chat
- ChatGPT Work or Chat? 5 Questions to Help You Choose
- ChatGPT Work vs Chat: Hands-On Test Results
- How to Use ChatGPT Work on Web, Desktop, iPhone, and Android
- ChatGPT Work Pricing, Plans, Models, and Usage Limits
- ChatGPT Work FAQ
- Conclusion: Use Chat by Default and Work When the Workflow Justifies It
- Sources and References
What Is ChatGPT Work? Key Differences from Chat
ChatGPT Work is a task-completion mode inside ChatGPT for longer, more involved work. It can research and analyze information, work with files and connected apps, and create reviewable documents, spreadsheets, presentations, reports, and Sites. Regular Chat remains the default for questions, discussion, brainstorming, short drafts, and straightforward tasks. The distinction is about workflow structure — not a guarantee that Work will produce a better or more accurate result.
OpenAI’s official overview of ChatGPT Work and Codex positions Chat as the place for questions and quick conversational help, while Work is intended for longer research and finished materials. Work also lets you follow progress, answer questions, change direction, and approve important actions while a task is running. The figure below compares these typical routes while recognizing that both modes can create editable files.
- Multiple sources must be compared or reconciled
- An editable DOCX deliverable is required
- Visible checkpoints or validation steps matter
File creation alone does not separate the two modes. In this single hands-on test, regular Chat read the same three official source documents and also produced an editable report. Work’s practical advantage was visibility: source review, drafting, word-count verification, and layout checks were easier to follow as stages within one managed task. This describes what happened in this test, not a general guarantee of better quality or accuracy.
The practical rule used in this article is to start with Chat by default. Consider Work when at least two of these conditions apply: multiple sources must be compared or reconciled, an editable DOCX deliverable is required, or progress checkpoints and validation steps need to remain visible. The next section expands this rule into five practical criteria for choosing between the two modes.
ChatGPT Work or Chat? 5 Questions to Help You Choose
The selector below starts with regular Chat as the default. Q1 asks whether one response or straightforward output can complete the job. If the answer is yes, stay in Chat. If the answer is no or unclear, Q2–Q5 test three conditions that may justify moving to Work: multiple sources that must be compared or reconciled, an editable DOCX deliverable, and visible checkpoints or validation steps.
The five questions are not five equal points. Q1 is the initial routing question, while a Yes to both Q2 and Q3 counts as one source-reconciliation condition. Q4 and Q5 count as the other two conditions. Consider Work only when at least two of the three conditions apply. A long prompt or several attached files does not automatically make Work necessary. The figure shows the complete decision process, and the sections that follow explain how each condition applies to real tasks.
Use Chat When One Response Is Enough
Chat is the default when the job is complete once you have received and reviewed one response or straightforward output — for example, a definition, brainstorm, short rewrite, concise summary, or draft you plan to finish yourself. The deciding factor is not how important the task feels or how detailed the prompt looks. It is whether additional stages need to remain visible and connected inside the same managed workflow.
In our July 16, 2026 English-language test, Chat and Work received the same notes and instructions for a 70–100-word progress update. Both produced a compliant update in the first response, and neither required a follow-up question or an additional work stage. In this single run, Work did not show a meaningful practical advantage for this type of short, self-contained task.
Before moving to Work, ask what must happen after the first output arrives. If you only need to read it, make a light edit, and use it, stay in Chat. If the task must also reconcile multiple sources, produce an editable DOCX, or preserve visible checkpoints and validation, continue through the remaining criteria. Consider Work only when at least two of those three conditions apply.
Consider Work When Multiple Sources Must Be Reconciled
Multiple sources become a reason to consider Work when they must be compared, reconciled, and carried into later stages. Attaching several files is not enough by itself. The relevant condition is whether differences, conflicts, missing details, and changing information must be resolved before the final output can be completed.
Multiple sources alone do not make Work necessary. If you only need separate summaries of three documents, regular Chat may still be the simpler choice. In the selector above, the source condition applies only when the task both uses multiple sources and requires them to be compared or reconciled. Even then, it counts as one of the three Work conditions rather than an automatic reason to switch modes.
In our July 16, 2026 English-language test, Chat and Work read the same three official OpenAI documents and produced editable reports from the same instructions. This confirmed that reading multiple files and creating a document are not exclusive to Work. Work’s practical advantage in this single run was visibility: source review, cross-document checking, drafting, word-count verification, and layout review were easier to follow as connected stages within one managed task.
Treat source reconciliation as the first Work condition, then check the other two: whether an editable DOCX is required and whether checkpoints or validation steps need to remain visible. If source reconciliation is the only condition that applies, Chat may still be enough. If it appears alongside at least one of the other conditions, Work becomes the stronger choice. Whichever mode you use, the final factual, structural, and visual review remains your responsibility.
Consider Work for Documents, Spreadsheets, and Presentations
Creating a file does not automatically make Work the right choice. Regular Chat can also produce editable outputs, as it did in our test. Work becomes a stronger candidate when the deliverable’s structure — such as sections, tables, formulas, slide order, citations, or formatting — must be developed and checked as part of a visible, connected workflow.
OpenAI states that Work can create or edit documents, spreadsheets, presentations, reports, and analyses. However, availability depends on the plan, workspace settings, file type, connected app, and surface being used (checked July 16, 2026). Native Google Docs, Sheets, and Slides require the relevant Google Workspace app to be enabled. Directly inspecting or updating an open Microsoft Excel workbook uses Codex with the ChatGPT for Excel add-in, while PowerPoint was not included in the Work desktop flow at launch.
In our July 16, 2026 DOCX test, Chat produced a compact two-page document, while Work produced a four-page document that separated pricing and plan information, confirmed versus changing conditions, points requiring verification, and source references into distinct sections. More pages do not automatically indicate a better result. In this single comparison, the additional separation made the requested sections easier to locate and review.
For the decision rule used in this article, requiring an editable DOCX counts as one of the three Work conditions; it does not justify switching modes by itself. If you mainly need text, an outline, or a straightforward file that you will place into your own template, start with Chat. Consider Work when the editable DOCX requirement appears alongside source reconciliation or a need for visible checkpoints and validation. Spreadsheet and presentation support is officially documented, but those formats were not directly compared in this hands-on test. Whichever mode you use, verify facts, calculations, citations, formulas, formatting, and sharing permissions before relying on the file.
Consider Work When You Need Checkpoints or Validation
Choose Work when a change must be carried through several connected sections and you want the affected areas and validation steps to remain visible. Chat can also revise a completed file, but Work becomes more useful when reviewing how the revision is being applied is part of the job—not just receiving the updated deliverable.
OpenAI describes Work as allowing users to follow progress, answer questions, redirect the task, and approve actions when required. This matters when one changed decision should propagate into the conclusion, comparison table, recommendations, source handling, and final checklist. If only the next paragraph changes, a normal follow-up in Chat may be enough; if the decision changes the rest of the workflow, visibility becomes more valuable.
Not every revision needs Work. If the change is narrow and you can verify the finished file directly, Chat is usually the simpler starting point. A practical rule is to use Work when a missed downstream change would be difficult to detect by reviewing the final file alone.
To test that distinction, we first completed the same three-source memo in both modes and then sent each one the same follow-up revision. The comparison below shows what remained visible while Chat and Work revised their respective editable DOCX files.
Both modes received the same follow-up instruction: Change the memo’s decision rule so that Chat is the default choice, and recommend Work only when at least two of three conditions apply: multiple sources must be reconciled, an editable DOCX must be produced, or progress and validation steps must remain visible. The audience, three official OpenAI source files, and required editable format were kept unchanged.
In this single July 16, 2026 test, both modes revised the memo and returned an editable DOCX. Chat showed a compact revision flow, completed the follow-up in 8 minutes 50 seconds, and returned a two-page portrait document. Work took 26 minutes 26 seconds and returned a four-page mixed-orientation document, while explicitly showing the affected sections, consistency checking, validation, and file replacement steps.
The observed difference was workflow visibility rather than whether the revision succeeded. Work made it easier to see where the new decision rule was being propagated, while Chat reached an updated deliverable through a shorter visible process. The longer Work run does not by itself prove greater accuracy, and the additional pages do not automatically make its result better.
This was one run per mode, not an average-performance benchmark. Chat’s displayed High setting and Work’s GPT-5.6 Sol with High effort are not identical controls, so the completion times should not be treated as a controlled speed comparison. Use Work when visible impact analysis and validation justify the additional time; use Chat when you need a direct revision and can inspect the finished file yourself.
Not Sure? Clarify the Task in Chat First
If you are unsure which mode fits, stay in Chat while you clarify the task. Use the conversation to define the goal, identify the required sources, choose the output format, resolve missing information, and write clear completion criteria. Then apply the three-condition rule. Move to Work only if at least two conditions apply; otherwise, complete the task in Chat.
Do not assume that moving from Chat to Work will carry every instruction, attachment, decision, and unresolved question into the new task. Restate the agreed direction, attach the required files again, and identify anything that must remain unchanged. This is especially important when changing surfaces. OpenAI states that cloud Work conversations do not appear in desktop Work at launch, while desktop Work threads and local files remain on that computer. Treat a move between Chat, cloud Work, and desktop Work as an explicit handoff.
In our July 16, 2026 English-language comparison, Chat and Work were started as fresh conversations. We used the same account, prompt, source files, output requirements, and completion criteria in both modes instead of relying on a direct transfer. Memory was enabled on the account, so saved preferences may still have influenced both outputs. This was therefore a controlled comparison where practical, but not a perfectly isolated benchmark.
The purpose of this two-stage method is to prevent an unclear request from becoming the wrong deliverable. Chat reduces ambiguity; the three-condition rule determines whether Work is justified. If the clarified task requires at least two of the following — source reconciliation, an editable DOCX, or visible checkpoints and validation — move it to Work with the handoff brief above. If fewer than two conditions apply, there is no need to leave Chat.
ChatGPT Work vs Chat: Hands-On Test Results
On July 16, 2026, we gave Chat and Work two matched English-language tasks: a 70–100-word progress update and a 900–1,200-word editable DOCX decision memo based on the same three official OpenAI sources. We matched the account, model, prompts, source files, output requirements, and completion criteria where practical. The goal was not to identify a universal winner, but to observe whether and where the workflows, visible review stages, and deliverable structures differed.
Each mode completed each condition once, so the results describe these specific runs rather than average performance. We checked first-response compliance, missing requirements, follow-up needs, workflow visibility, and whether the final file opened, edited, and retained its structure cleanly. Processing time and usage are included as supporting observations, not as standalone measures of quality. The sections below separate the setup, short task, multi-source task, and evidence limits so that the conclusions remain proportional to what this comparison can establish.
Test Setup and Evaluation Criteria
The test-environment box near the beginning of this article records the account, surface, model, displayed settings, and limitations. This section focuses on what each mode was asked to do and how the resulting outputs were evaluated.
On July 16, 2026, we ran two matched English-language tasks. For each task, Chat and Work received the same prompt, source material, output language, and completion criteria in fresh conversations. Each task was run once per mode. GPT-5.6 Sol was used in both modes, but the controls were presented differently: Chat displayed High, while Work displayed High effort and Standard speed. Because the interfaces did not expose the settings in the same format, the runs should not be treated as having identical execution settings.
The first task asked each mode to turn the same notes into a 70–100-word English progress update for a manager. The response had to state the current progress in the first sentence, avoid bullet points, maintain a professional but not overly formal tone, add no unsupported facts, and output only the completed update.
The second task gave both modes the same local PDF copies of three official OpenAI pages and asked for a 900–1,200-word evidence-based decision memo as an editable DOCX. The memo had to include an executive conclusion, a plan-and-access comparison, confirmed facts separated from changeable conditions, practical Chat-versus-Work guidance, a source map, and a five-item pre-publication checklist.
For the short task, we checked first-response compliance, word count, unsupported additions, follow-up needs, and whether the finished paragraph was usable. For the multi-source task, we also checked source reconciliation, required-section coverage, DOCX editability, page structure, and how clearly checkpoints and validation steps remained visible.
Processing time is included as supporting information, not as a standalone measure of quality. We did not measure the additional time a person would need to fact-check, revise, and prepare either document for publication. The separate usage observation covered one mixed session involving several activities, so its before-and-after percentage cannot be assigned to either matched task in isolation.
Short Writing Test: No Meaningful Difference
In this one-run short writing test, Chat and Work produced equally usable progress updates; the practical difference was how the result was presented, not whether the task was completed. Both briefly showed “Thinking” and then returned a finished paragraph without asking a follow-up question or exposing a separate planning workflow.
Both modes received the same English prompt and the same source notes. The task was to rewrite the notes as a 70–100-word progress update to a manager, state the current progress in the first sentence, avoid bullet points, use a professional but not overly formal tone, add no unsupported facts, and output only the completed update.
The notes stated that data collection was complete, two charts still needed to be created, the internal review would take place on Friday, and publication was planned for the following Monday. We ran the test on ChatGPT Plus using the web interface in English on July 16, 2026, with Chat set to High and Work set to GPT-5.6 Sol · High; no tools or external sources were used.
Chat returned 71 words, while Work returned 77 words, so both stayed within the requested 70–100-word range. Chat presented its answer in an editable card, whereas Work returned a standard message; this was a visible interface difference, not evidence of a meaningful difference in answer quality.
The wording was not identical. Chat produced the more compact update, while Work used its six additional words to explain the remaining charts and short review window in slightly greater detail. Both conveyed the current status, schedule, remaining work, and timing concern without using bullet points.
This was one run per surface, not an average-performance test. Chat and Work also displayed different model controls, so the result does not isolate the effect of the interface from every model-related variable, nor does it prove that both modes will perform equally on all short writing tasks.
For a single paragraph that can be reviewed immediately, Chat remains the simpler default. Work completed the task successfully, but this test found no added workflow benefit large enough to offset opening Work for such a small, one-step deliverable.
Multi-Source Report Test: Work Was Easier to Track
In this one-run multi-source report test, Work made the workflow easier to track, although both Chat and Work successfully turned the same three official OpenAI PDFs into editable DOCX files. The practical difference was not whether either mode could create the requested deliverable, but how much of the source handling, progress, and validation process remained visible while it was being produced.
The source set consisted of local PDF copies of three official OpenAI pages: the ChatGPT plans and pricing page, the ChatGPT download page, and the “What is ChatGPT Plus?” Help Center article. Both modes were asked to use only those three files—without web browsing or outside information—to create a 900–1,200-word evidence-based decision memo for an individual ChatGPT Plus subscriber deciding when and where to use ChatGPT Work.
The editable memo had to include an executive conclusion, a plan-and-access comparison, confirmed facts separated from conditions that may change, practical Chat-versus-Work guidance, a source map, and a five-item pre-publication verification checklist. We ran the test on ChatGPT Plus using the English web interface on July 16, 2026, with Chat set to High and Work set to GPT-5.6 Sol · High; the prompts and source files matched, but the displayed model controls were not fully identical.
Chat completed its memo in 7 minutes and returned 1,192 words, while showing the main stages as a compact activity list. Work completed its memo in 19 minutes 20 seconds and returned 1,144 words; during that longer run, it showed how the PDFs were being assigned source roles, how the memo would be structured, and what was being checked before delivery.
Work also kept its source map, validation checklist, Sources panel, and editable DOCX output visible as connected parts of the task. Neither mode required an additional reply from the user, so the distinction in this run was workflow visibility rather than fewer questions or fewer instructions.
The longer Work process should not be treated as evidence that Work is always slower, and the extra activity labels do not by themselves prove that its claims or final document were more accurate. They show what the interface exposed as part of the workflow; the underlying prices, model names, citations, recommendations, and layout still need to be checked in the finished file.
Each surface was tested once, and the displayed model controls were not identical, so these times are observations from this task rather than an average-performance benchmark. What this run supports more directly is that Work made a long, source-dependent task easier to monitor while it was still in progress.
Both modes delivered the requested editable memo and included the comparison, source information, practical guidance, and verification checklist. Chat packaged its 1,192 words into two portrait pages, while Work spread its 1,144 words across four pages using both portrait and landscape orientations.
Chat placed the conclusion and plan comparison on the first page, then combined the guidance, sources, and checklist on the second. Work used a portrait opening page for its evidence and conclusion, a landscape page for the plan comparison, another landscape page for the guidance and source map, and a final page for the verification checklist.
Chat therefore produced the more compact document, even though it contained slightly more words. Work used fewer words but gave major requirements more visual separation, making the comparison table, source map, and final checks easier to locate individually.
A shorter file is not automatically better, and a longer file is not automatically more complete. In this test, the difference was how the same type of evidence was packaged for review: Chat favored density, while Work favored visible separation and document structure.
These page counts describe the two files generated in this single run; they are not fixed Chat or Work behavior. The final choice therefore depends on whether you value a compact memo that is faster to scan or a more separated layout that makes each required section easier to inspect.
| Required element | Chat | Work | What was checked |
|---|---|---|---|
| Editable DOCX deliverable | Met | Met | Both modes returned a DOCX that opened as an editable document. |
| Overall length of 900–1,200 words | 1,192 | 1,144 | Both files stayed within the requested English word-count range. |
| Executive conclusion | Met | Met | Both files opened with a distinct conclusion for an individual ChatGPT Plus subscriber. |
| Plan and access comparison table | Met | Met | Both compared plans, Work availability, supported surfaces, model information, and important caveats. |
| Confirmed facts separated from variable conditions | Met | Met | Chat used a “Directly confirmed facts versus variable conditions” section. Work separated “What is directly confirmed” from “What may change.” |
| Practical Chat-versus-Work guidance | Met | Met | Both included a dedicated section explaining when an individual Plus subscriber might choose regular Chat or Work. |
| Source map with all three filenames | Met | Met | Both mapped memo sections to 01-chatgpt-pricing.pdf, 02-chatgpt-download.pdf, and 03-chatgpt-plus-help.pdf. |
| Five-item pre-publication verification checklist | 5 items | 5 items | Both ended with exactly five checks covering current access, models, pricing or limits, supported surfaces, and other changeable conditions. |
How to Use ChatGPT Work on Web, Desktop, iPhone, and Android
ChatGPT Work is available on the web, in the desktop app, and on supported iPhone and Android accounts, but file access, conversation history, and available controls are not identical across those surfaces. Web and mobile Work run in the cloud, while the desktop app can use permitted local files and supported desktop applications. Cloud Work conversations do not appear in desktop Work at launch, and desktop threads and local files remain on that computer. Availability and interface details may depend on your plan, workspace settings, app version, and rollout status (checked July 16, 2026).
The practical rule is to choose the surface by where the required files live and where you plan to review the task. Use Web for browser-based work involving uploaded files or permitted cloud services. Use Desktop when the task depends on files or supported applications on your computer. Use iPhone or Android to start, monitor, or review cloud Work while away from your desk. This article tested Web, macOS, and iPhone hands-on; Windows and Android guidance is based on OpenAI’s official documentation. The figure below helps you choose a surface, while the four H3 sections that follow provide the separate opening steps, screenshots, and limitations for each one.
- Starting a Work task from a browser
- Uploading selected source files
- Continuing Work across supported cloud surfaces
- Selecting local files
- Working with supported desktop apps
- Keeping local project files together
- Selecting Work from the iPhone mode menu
- Reviewing progress away from your computer
- Answering questions or redirecting a task
- Selecting Work from the mobile mode menu
- Starting or continuing a cloud task
- Reviewing progress and responding to questions
How to Use ChatGPT Work on the Web
To start Work on the web, open ChatGPT, start a new conversation, and select Work from the Chat–Work switch at the top of the screen before sending your instructions. In the English web interface we tested, Work opened within the regular ChatGPT site rather than at a separate Work-specific URL.
- Open ChatGPT on the web and start a new conversation.
- Select
Workfrom the Chat–Work control at the top of the page. - Confirm that Work is selected and that the Work task-entry screen is displayed.
- Add only the files, project, plugin, or connected source that the task actually needs.
- Describe the intended outcome, permitted source material, constraints, and final deliverable.
- Submit the task, then review any question, permission request, or approval step if one appears.
A short topic alone is usually not enough for a delegated task. Before submitting it, specify what Work should produce, which evidence it may use, and the format in which the result should be returned.
On my ChatGPT Plus account, the English web interface displayed Chat and Work as a two-option control at the top of the new-task screen on July 16, 2026. After Work was selected, the page changed to the Work entry screen, where the task and deliverable could be described before submission.
The left side of the image above shows the web procedure; the iPhone procedure is included on the right for comparison and is explained separately below. The control’s position, wording, and availability may vary with the plan, workspace, account, or rollout status, so this screenshot documents the interface observed on that date rather than guaranteeing a permanent layout for every account.
Web Work can use files that you explicitly upload and any connected sources made available to the task, but it does not receive unrestricted access to files stored on your computer. For a local file, use the upload control and select the specific document you want to provide rather than assuming Work can search the rest of the device.
Projects, plugins, and connected services may expose different information or actions depending on their permissions. Add only what the task needs, check which account or workspace is connected, and review any read, write, or approval request before allowing an external action to continue.
A useful web brief defines the outcome, permitted source material, final format, constraints, and acceptance checks. If some information must remain unchanged—or if Work must not browse the web, use outside sources, or modify a connected service—state those boundaries before starting the task.
This does not make every request error-free, but it gives Work a clearer completion target. The final file, citations, calculations, and any external changes should still be reviewed before you accept or share the result.
How to Use ChatGPT Work on Desktop (macOS and Windows)
To start Work in the desktop app, open the current ChatGPT app, switch from Chat to Work, and attach only the local files or connected sources needed for the task. The desktop app is especially useful when the work depends on documents stored on your computer or a desktop-specific working environment.
- Install or open the current ChatGPT desktop app for macOS or Windows.
- Start a new task and select Work from the Chat–Work control.
- Choose a project if the task belongs to an existing workspace.
- Attach only the local files the task needs, or select an available connected source.
- Review any permission or approval request before allowing access or an external action.
- State the intended deliverable, permitted sources, constraints, and completion checks.
File access and task instructions are separate decisions. Selecting the right files controls the available evidence; the brief still needs to explain what Work should produce and how you will judge whether the result is complete.
In our July 16, 2026 desktop test, we attached three local Markdown files and instructed Work to identify which files were available without summarizing, quoting, or editing their contents. After 12 seconds, Work returned test-a.md, test-b.md, and test-c.md; the same three filenames also appeared in the Sources panel.
This confirms that the tested Work task recognized the three files we deliberately attached. It does not establish a general boundary for local-file access, prove that only filename metadata was processed, or show what would happen under different permissions. The prompt did not request file contents, so this test should be treated as evidence of file selection and recognition—not a privacy or access-scope audit.
The desktop app can use files you explicitly attach and may offer additional app or system connections when supported, but it should not be treated as unrestricted access to your computer or as an offline version of ChatGPT. Start with the smallest practical source set. If Work requests permission to read, change, move, or share information, check both the target and the scope before approving it.
The current desktop experience can expose Chat, Work, and other agent surfaces in the same application. If Work is missing, first confirm that the desktop app is current and that you are signed in to the intended account or workspace. An older installation, workspace restriction, staged rollout, or plan difference can affect what appears.
The mode names, locations, and available connections can vary by plan, account, workspace, operating system, and rollout status. The interface and file workflow shown above were observed in the English ChatGPT desktop app on July 16, 2026.
Do not assume that every local file, file path, permission, or desktop-only connection will remain available after you move to another device or surface. A conversation may still appear under the same account while its original local source is unavailable from the new environment.
Before switching devices, confirm where the latest output was saved, whether it opens correctly, and whether the next environment has access to every required source. For an important deliverable, download or copy the finished file to a location you control instead of relying only on the task view.
How to Use ChatGPT Work on iPhone and Android
On iPhone and Android, open the ChatGPT app, tap the dropdown showing Chat or Work at the top of the screen, and select Work. OpenAI currently documents one shared mobile starting path rather than separate iPhone and Android procedures. This article confirmed the interface hands-on on an iPhone 16 Pro; the Android instructions below are based on official documentation rather than a direct Android test. OpenAI’s current Work help was checked on July 16, 2026.
- Open the official ChatGPT mobile app and sign in.
- Tap the dropdown at the top of the screen showing
ChatorWork. - Select
Work. - Start a new conversation or open an existing project.
- Add the files and context that the task is permitted to use.
- Describe the task, required deliverable, constraints, and review criteria.
On an iPhone 16 Pro using a personal ChatGPT Plus account, Work appeared in the top mode dropdown on July 16, 2026. After selecting Work, the task could be entered through the mobile composer, with files and context added from sources accessible to the phone. The iPhone screenshot used in this article records that observed interface; the exact placement, wording, and available controls may change with the app version and rollout.
This article does not claim hands-on verification of the Android Work interface. OpenAI’s current Help Center describes one mobile dropdown for switching between Chat and Work rather than a separate Android-specific Work procedure. Android users should therefore confirm the live interface in the official app. If the dropdown appears in a different position or Work is missing, check the current ChatGPT Work and Codex help article, your plan eligibility, workspace settings, and app version rather than relying only on the iPhone screenshot.
Work on iPhone and Android runs in the cloud. It can use files intentionally added to the mobile task and permitted connected services, but it cannot directly access files stored only on another computer. A Work conversation created on web or mobile can continue across supported cloud surfaces. OpenAI states that cloud Work conversations do not appear in desktop Work at launch, while desktop threads and local files remain on that computer.
For individual plans, OpenAI’s pricing page listed Work on desktop, web, and mobile for Plus and Pro, while Free and Go had limited desktop-only Work access when checked on July 16, 2026. Availability may still depend on the account, workspace settings, app version, and rollout status. If Work is missing, check the account’s current eligibility and the latest official documentation before assuming that the app is malfunctioning.
ChatGPT Work Not Showing? 6 Things to Check
If ChatGPT Work is missing, check the correct interface location, account and workspace, plan eligibility, rollout status, and app version — in that order. A missing Work option does not necessarily mean that ChatGPT is malfunctioning.
First, check the selector for the surface you are using. On web and mobile, open a new conversation and use the dropdown at the top of the screen showing Chat or Work. In the desktop app, Work and Codex are selected from the mode switcher near the top left, while regular Chat is opened through Quick chat. Checking the wrong selector can make an available mode appear to be missing.
Next, confirm that you are signed in to the intended OpenAI account and workspace. OpenAI describes Work as a gradual rollout to eligible accounts. For individual plans, the pricing page listed Work on desktop, web, and mobile for Plus and Pro, while Free and Go had limited desktop-only access when checked on July 16, 2026. Check the current ChatGPT Work documentation before changing plans, because eligibility and rollout details may change.
If you use a managed Business, Enterprise, or Edu workspace, switch between your personal and managed workspaces and check again. Feature visibility can differ by workspace, and administrator settings may control whether Work is enabled. If Work appears in your personal workspace but not in the managed one, ask the workspace administrator before troubleshooting the device.
On iPhone or Android, confirm that you installed the official ChatGPT app and apply any available update. Android users can find the installed version under Sidebar → profile icon → About, according to OpenAI’s Android App FAQ. Updating the app may restore a newer interface, but it cannot grant Work access to an account or workspace that is not yet eligible.
For the desktop version, confirm that you are using the current ChatGPT desktop app and compare the same account and workspace on the web. If Work appears on the web but not on desktop, check the desktop app version, plan coverage, workspace controls, and rollout status. Remember that cloud Work and desktop Work are separate surfaces at launch, so their conversations and local-file access do not fully match.
- Check the correct Chat/Work selector for web, mobile, or desktop.
- Confirm the OpenAI account and workspace currently selected.
- Check the latest plan eligibility and rollout information in OpenAI’s documentation.
- Ask the workspace administrator if you use a managed Business, Enterprise, or Edu account.
- Update the official mobile or desktop app, then compare the same account and workspace on the web.
- If access still appears to be missing, contact OpenAI Support with your plan, workspace type, device, operating system, app version, and a screenshot of the missing selector.
ChatGPT Work Pricing, Plans, Models, and Usage Limits
OpenAI’s pricing page lists ChatGPT Work as a feature within ChatGPT plans rather than showing a separate Work-only price. That does not mean every account receives the same access. Plan coverage, supported surfaces, available models and controls, and included usage can vary by plan, workspace, account, and rollout status, so each layer needs to be checked separately.
The most reliable order is to confirm four things: the current plan listing and price; the surfaces and workspace access included for that plan; the model and controls displayed inside Work; and the remaining allowance, reset time, and shared features shown on the account’s Usage screen. The figure below separates those checks so that “listed with my plan” is not mistaken for “available on every surface with unlimited usage”.
- Free / Go — Limited desktop access listed
- Plus / Pro — Desktop, web, and mobile access listed
- Managed workspaces — Check plan coverage, administrator settings, and rollout
Settings → Usage → Usage limits
Is ChatGPT Work Free? Pricing and Supported Plans
ChatGPT Work can be available without a separate Work subscription, including limited access on the Free plan, but free access is not the same as full access across every device. Work is included as a feature within eligible ChatGPT plans rather than sold as a standalone product with its own monthly charge.
When checked on July 16, 2026, OpenAI’s ChatGPT pricing page listed limited desktop-app access to Work for Free and Go. Plus and Pro were listed with Work on desktop, web, and mobile. Eligibility still does not guarantee that the same interface, models, or usage allowance will appear in every account at the same time.
The figure below summarizes the plan-and-surface information displayed on the official pricing page on that date. It should be read as a dated reference rather than a permanent plan guarantee, because plan names, included features, and availability can change.
If you search for “ChatGPT Work pricing,” the important distinction is that you pay for the underlying ChatGPT plan, not an additional Work subscription. The plan determines which Work surfaces and level of access may be included. Check the current official ChatGPT pricing page before subscribing, because prices, billing options, and included features may vary by plan, region, and date.
For Business, Enterprise, or Edu, plan eligibility is only part of the answer. The selected workspace may also apply administrator controls, security policies, contractual settings, regional restrictions, or staged availability. If Work appears in a personal workspace but not in a managed one, upgrading or reinstalling the app may not resolve the actual cause.
In my hands-on check on July 16, 2026, Work was available in a personal ChatGPT Plus account in Japan on the web, iPhone, and the current desktop app. This confirms those three surfaces for the tested account and date; it does not prove simultaneous availability for every Plus subscriber, country, or managed workspace.
For most individual users, the practical decision is therefore not whether to purchase Work separately. It is whether the Work access included with the plan and devices they already use is broad enough for their tasks—and whether its usage limits justify moving a task out of regular Chat.
Which Models Are Available in ChatGPT Work?
The models you can use in ChatGPT Work are the models currently listed in your own Work settings. The available names and controls can differ by plan, account, workspace, device, and rollout, so a model shown in another user’s screenshot is not a permanent compatibility guarantee.
In my personal ChatGPT Plus account on July 16, 2026, the English web interface listed GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, and GPT-5.5. The tested interface treated the selected model, effort level, and response speed as three separate controls rather than one combined quality setting.
The effort menu offered Light, Medium, High, Extra High, and Max, while the speed menu offered Standard and Fast. The interface described Fast as faster but using more of the available allowance, so changing speed may affect consumption as well as waiting time.
For the hands-on tests in this article, Chat used the interface’s High setting, while Work used GPT-5.6 Sol with High effort and Standard speed. Although both surfaces exposed GPT-5.6 Sol, their visible controls were not presented in exactly the same way.
The label “High” in Chat should not be assumed to equal Work’s High effort setting, and Work’s separate Standard speed setting adds another variable. The results therefore compare the two user-facing modes under the closest practical settings available to this account—not a laboratory benchmark with identical hidden configuration.
The important point is that GPT-5.6 Sol was not exclusive to Work in the tested account. Regular Chat could also expose GPT-5.6 Sol, so the reason to choose Work is its task workflow, source handling, checkpoints, and deliverable creation—not the model name alone.
When several models and effort levels are available, choose the configuration according to the cost of getting the task wrong. A short rewrite or simple formatting request usually does not need the highest effort setting. A multi-source report involving conflicting evidence, calculations, file creation, or verification may justify more effort.
Speed is a separate trade-off. Fast can reduce waiting time, but the tested interface warned that it uses more allowance. If the task is not urgent, Standard is the safer starting point; move to Fast only when the time saved matters more than the additional consumption.
The highest model, effort, and speed settings should not become an automatic default. Start with a proportionate configuration, inspect the result, and increase effort only when the task’s complexity or risk calls for it.
- The full model name displayed in Work
- The selected effort level
- The response-speed setting
- Whether each setting could be changed manually
Recording all four items makes later comparisons easier to reproduce. It also prevents a model, effort, or speed change from being mistaken for evidence that Chat and Work themselves produced different results.
Model names, available effort levels, and speed controls are highly changeable. Treat the list shown above as an observation from one personal Plus account and the English web interface on July 16, 2026—not as a permanent model-support list.
For OpenAI’s current description of the feature, check the official ChatGPT Work documentation. For the models you can actually select right now, the settings displayed in your own Work interface are the final account-level reference.
ChatGPT Work Usage Limits: How to Check Your Allowance
ChatGPT Work has no publicly stated universal limit expressed as one fixed number of tasks per day or month. Available usage can depend on the plan, selected model, reasoning level, files, tools, and the amount of processing required by the task (checked July 16, 2026).
A short rewrite and a multi-source report do not necessarily use the same amount of the included allowance. A task count reported by another user—or a limit published for a different ChatGPT feature—should therefore not be treated as your own Work limit. The remaining percentage, reset time, and shared features displayed in your account are the most relevant starting points.
On the ChatGPT Plus account used for this article, the displayed weekly usage remaining changed from 96% at 2:32 PM to 82% at 3:39 PM on July 16, 2026—a decrease of 14 percentage points over about 1 hour 7 minutes.
This was a mixed session rather than one isolated task. It included creating and revising two short DOCX files of roughly two to three pages each, while fixed website pages—including the terms, home, and about pages—were also edited in parallel. The Usage screen displayed a reset time of July 23, 2026 at 1:53 PM.
This observed 14-percentage-point decrease does not mean Work always uses 14 points per hour, or that the DOCX work alone caused the change. Several activities were performed during the same session, the interface displayed whole-number percentages, and the amount used can vary with the model, reasoning level, source volume, connected tools, and deliverable complexity. This observation therefore cannot establish a reliable per-task or per-hour consumption rate.
The Usage screen on this account also stated that the allowance was shared across Codex, Work, Workspace Agents, and ChatGPT for Excel, while regular Chat conversations were not included in that particular shared allowance. This means the before-and-after result cannot prove that Work alone accounted for the entire change. Check OpenAI’s credit and flexible-usage documentation for the current rules, since the features sharing an allowance and the available credit system may change.
- Open ChatGPT and select your profile menu.
- Go to
Settings → Usage → Usage limits. - Record the date, time, remaining percentage, reset time, and features sharing the allowance.
- Record the selected model, reasoning level, speed setting, source volume, and task scope.
- Avoid unrelated use of other features sharing the allowance during the test period.
- Run one representative Work task, then return to the same screen and compare the result.
- Repeat comparable tests before treating the observed change as a useful range.
Measure a task that resembles your real work, but isolate it from unrelated activity as much as practical. A very short prompt may not produce a visible change on a whole-number percentage display, while a mixed session may produce a larger change that cannot be assigned reliably to any one task.
For a more useful estimate, repeat comparable tasks under similar model, reasoning, speed, source, and deliverable conditions. Treat the results as an observed range rather than a guaranteed consumption rate.
If your included allowance runs out, additional credits may be available for eligible accounts. Buying credits changes how much supported usage can continue—it does not turn Work into an unlimited service. Check the price and eligible features displayed in your own account before purchasing.
The practical question is therefore not “How many times can everyone use Work?” but “Does my current allowance support this type of workload, and does the resulting quality and visible workflow justify the additional time and usage?”
ChatGPT Work FAQ
This FAQ answers the practical questions that may remain after comparing regular Chat and ChatGPT Work, including pricing and availability, usage limits, mobile access, accuracy, file handling, syncing, human review, and the difference from Codex. Availability and plan details were checked against OpenAI’s official documentation on July 16, 2026 and may change. Open only the questions relevant to you.
What is the difference between ChatGPT Work and regular Chat?
Regular Chat is designed for questions and conversational help, while ChatGPT Work is designed to carry longer, multi-step tasks through to reviewable deliverables. Work is not automatically a more powerful version of Chat; its main difference is the workflow used to connect research, analysis, file creation, checkpoints, and review.
In this article’s tests, both modes completed a short writing task without a meaningful practical difference. Work became more useful when three sources, multiple requirements, and an editable report had to be managed as one connected task.
Is ChatGPT Work free?
ChatGPT Work is not sold as a separate subscription. Free and Go may receive limited access through the desktop app, while OpenAI lists broader Work access across desktop, web, and mobile for Plus and Pro.
Availability in managed workspaces can also depend on the organization’s plan, administrator settings, supported region, and rollout status. Plan coverage may change, so check the current ChatGPT pricing page before subscribing or changing plans (checked July 16, 2026).
How many times can I use ChatGPT Work?
OpenAI does not publish one fixed daily or monthly ChatGPT Work limit that applies to every user and task. Available usage can depend on the plan, selected model, reasoning level, files, tools, and the processing required by the requested deliverable.
On the Plus account tested for this article, the displayed weekly usage remaining changed from 96% at 2:32 PM to 82% at 3:39 PM on July 16, 2026 — a 14-percentage-point change over about 1 hour 7 minutes. The session mixed several activities, including creating and revising two short DOCX files and editing website pages in parallel.
This single observation does not establish a fixed per-task or per-hour consumption rate. The change also cannot be attributed to one task or to Work alone because the displayed allowance was shared across Codex, Work, Workspace Agents, and ChatGPT for Excel. Check Settings → Usage → Usage limits for your own remaining percentage, reset time, and shared features.
Why is ChatGPT Work not showing in my account?
ChatGPT Work may be missing because your plan, device, workspace, region, or account has not received access yet. OpenAI describes Work as a gradual rollout, so an absent option does not necessarily indicate an app error.
Check the Chat/Work mode selector, selected account and workspace, current app version, and official eligibility information before reinstalling the app or changing plans. Managed-workspace users should also ask their administrator whether Work is enabled (checked July 16, 2026).
Can I use ChatGPT Work on iPhone and Android?
ChatGPT Work is available through the official iPhone and Android apps for eligible paid accounts. OpenAI’s pricing information listed Free and Go access as limited to the desktop app, while Plus and Pro received broader desktop, web, and mobile access as of July 16, 2026.
On mobile, open a new conversation and use the dropdown at the top of the screen to switch from Chat to Work. This article confirms that procedure on an iPhone 16 Pro; Android availability and the starting procedure are based on OpenAI’s official documentation rather than a hands-on Android test.
Is ChatGPT Work better or more accurate than Chat?
No. ChatGPT Work should not be treated as automatically better or more accurate than regular Chat. Accuracy still depends on the selected model, source quality, instructions, task complexity, and human review.
Work can make research, production, and checking stages easier to follow, but visible progress does not prove that every claim or calculation is correct. In this article’s short-writing test, Work did not produce a meaningful practical advantage over Chat.
Can ChatGPT Work edit documents, spreadsheets, presentations, or templates?
ChatGPT Work can create or edit supported files and can use an existing file or reusable template as a reference. Specify what should change and what must remain unchanged, including formulas, layout, branding, slide order, table structure, and document tone.
Native Google Docs, Sheets, and Slides require the relevant Google Workspace app to be connected and enabled. OpenAI also notes that directly editing an open Excel workbook uses Codex with ChatGPT for Excel, while PowerPoint was not included in the Work desktop flow at launch. Support can vary by file type, plan, workspace, and surface.
Can ChatGPT Work access files on my computer?
ChatGPT Work can access selected local files through the desktop app after you grant permission. Web and mobile Work cannot directly browse files stored only on your computer.
Open only the folder or project required for the task and avoid granting broader access than necessary. According to OpenAI’s Work documentation, local files and outputs remain on that computer unless you explicitly move or share them.
Can I move an existing Chat conversation into Work?
Not as a guaranteed complete handoff. Do not assume that switching from Chat to Work will automatically transfer every message, attachment, decision, and condition from the existing conversation.
The safer approach is to start a new Work task with a short handoff containing the objective, decisions already made, unresolved questions, required source files, constraints, and completion criteria. Cloud and desktop Work should also be treated separately because their conversation and local-file behavior can differ.
Where are ChatGPT Work files saved, and do they sync across devices?
Cloud Work files may be saved to Library where available, while local desktop outputs may remain only on the computer where they were created. Files created in Temporary Chat are not saved to Library.
Cloud Work conversations can continue across supported cloud surfaces, but cloud and desktop Work are not fully interchangeable at launch. Before changing devices, verify the filename, storage location, editable format, sharing permissions, and whether the conversation or file is visible on the destination device.
Can I stop ChatGPT Work or change direction while it is running?
Yes. ChatGPT Work lets you review progress, answer questions, add conditions, and redirect the task while it is running. However, a completed file change or approved external action may not always be automatically reversible.
When changing direction, state what should remain, what should change, which stage should be repeated, and what the revised deliverable must contain. Review the final output for content left over from the previous direction.
Can I share a file created by ChatGPT Work without reviewing it?
No. A file created by ChatGPT Work should be reviewed by a person before it is shared, published, or used for a decision. Visible checking stages do not guarantee that every fact, number, citation, formula, or layout element is correct.
Check these items before sharing:
- Source-to-claim correspondence
- Facts and calculations
- Required sections and conditions
- Citations and source links
- Editable format and page structure
- Broken characters and layout problems
- File access and sharing permissions
Work can reduce the effort required to reach a first deliverable, but responsibility for approving the final version remains with the user.
What is the difference between ChatGPT Work and Codex?
ChatGPT Work is intended for research and business deliverables, while Codex is designed for software development and technical changes. Choose Work when the final output is a report, document, spreadsheet, presentation, analysis, or similar business deliverable.
Choose Codex when the task involves source code, repositories, terminals, commands, tests, or configuration files. The deciding question is whether the final object being created or changed is a business deliverable or a software system.
Conclusion: Use Chat by Default and Work When the Workflow Justifies It
Start with regular Chat by default. Consider Work when at least two of these conditions apply: multiple sources must be compared or reconciled, an editable DOCX deliverable is required, or visible checkpoints and validation steps matter. File creation alone does not separate the modes, and Work is not an automatically better or more accurate version of Chat.
In the July 16, 2026 hands-on tests, Chat and Work both completed the short English writing task in their first response, with no meaningful practical difference. In the three-source report test, both produced editable DOCX files and included all eight requested deliverable elements. Work’s practical advantage was that source review, comparison, drafting, word-count verification, and layout checking were easier to follow as connected stages within one managed task.
That additional workflow came with trade-offs. Work took longer in these tests, and the account’s displayed weekly usage remaining changed from 96% to 82% during a separate mixed session lasting about 1 hour 7 minutes. Because that session included several activities and the allowance was shared across supported agentic features, the observed change cannot establish a fixed cost per task or per hour. The decision is therefore not whether Work is universally better, but whether its visible workflow and deliverable quality justify the additional time and usage for the task in front of you.
- Check the plan, surface, workspace, active model, and remaining usage before starting an important task.
- Review facts, calculations, citations, file structure, formatting, and permissions before sharing the result.
- Treat the processing-time and usage figures in this article as single-session observations, not fixed rates.
Once this routing becomes automatic, the more important question is how much of the workflow to delegate to AI and which decisions must remain under human review. The goal is not to choose the most advanced-looking mode, but to use the smallest workflow that can produce a reviewable result without giving up human responsibility for the final version.
Sources and References
Announcement1 source
Mode Selection & Availability2 sources
Pricing, Models & Usage3 sources
Deliverables, Storage & Desktop2 sources
External Apps & Android2 sources
Thank you for reading.
コメント