ChatGPT Free vs Plus

Keep ChatGPT Free if your work is mostly text: GPT-6 Luna chats are unlimited at $0. Pay $20 for Plus only when GPT-6 Sol, larger context, expanded work tools, or up to 3 hours of GPT-Live-1 Voice per rolling 24 hours save more than 12 to 48 minutes a month, depending on what your time is worth.

Which should you pick: ChatGPT Free or Plus?

Pick ChatGPT Free for everyday questions, drafting, summarizing, search, and low-stakes thinking. Pick Plus when a harder workflow needs GPT-6 Sol, more context, expanded access to files and research tools, hourly or event-triggered scheduled tasks, up to 3 hours of GPT-Live-1 Voice per rolling 24 hours, or ChatGPT Work beyond the desktop app.

Since October 7, 2026, OpenAI has been rolling out GPT-6 with Intelligent UI in Chat for Plus, Pro, Business, and Enterprise, with Free and Go following from October 8. Free and Go use GPT-6 Luna; the paid plans above use GPT-6 Sol. The Pro reasoning option remains GPT-6 Astra. Account availability can lag the rollout, and this release does not change the models powering Work or Codex.

The price and feature rows below were verified against OpenAI's live plan grid and current feature documentation on October 9, 2026. Prices are the listed US monthly rates; tool and model allowances remain subject to account availability.

The table needs one important distinction: unlimited text is not unlimited ChatGPT. On Free, a plain Luna conversation can continue while an image, upload, data-analysis job, GPT, or deep-research task is waiting for its own meter to reset. Plus expands those meters, but OpenAI still says limits can change with demand. Neither plan gives you a contractual amount of every tool.

Choose by the job that pays for the plan:

• A casual user or student doing ordinary text work should stay on Free. Luna can draft an email, explain a concept, summarize pasted text, and search the web. Paying $240 a year for Sol before Luna fails a meaningful job is buying capacity without a requirement.
• A senior operator should buy Plus when source files, longer context, deep research, or recurring deliverables are part of the week. The purchase is not “better AI” in the abstract. It is fewer broken handoffs between a document, a research pass, a decision, and a finished artifact.
• A funded founder should start on Free for exploration, then move to Plus when one board update, pricing analysis, hiring plan, or product brief repeatedly benefits from Sol. At a $100 hourly value, Plus only needs to recover 12 minutes per month. A single avoided rewrite can clear that threshold.
• A solo technical builder should buy Plus for broader Codex and Work access, not for API credits. OpenAI states that API usage is a separate bill. A Plus subscription does not fund an application running on the OpenAI API.
• A mid-market CTO should choose neither personal plan for governed company use. Free and Plus both lack an admin console, SAML SSO, unified billing, dedicated workspace controls, and a no-training-by-default business boundary. Plus is a stronger individual tool, not an organizational deployment plan.

This article narrows the decision to these two plans. For the full ladder from Free through Go, Pro, Business, and Enterprise, use the ChatGPT pricing breakdown. For the model-specific access question, read Is GPT-6 Luna free?. For the broader product verdict and alternatives, read the ChatGPT review.

ChatGPT Free is now the right default for everyday text

ChatGPT Free is the better purchase for anyone whose month is mostly ordinary text, because its subscription price is $0 and ordinary Luna text is now unlimited. That turns Free from a short trial into a durable default for a large class of work.

GPT-6 Luna powers the Free Chat experience as the rollout reaches accounts. It also supports Intelligent UI, the response format that can combine text, visuals, and interactive elements inside a conversation. A visual explanation or an in-chat calculator is therefore not, by itself, a reason to buy Plus. OpenAI’s Intelligent UI guide confirms that both Luna and Sol support it.

For everyday work, that is enough more often than the old Free-versus-Plus advice admits. A local-services owner can rewrite a customer reply, turn rough notes into an estimate template, or compare two scheduling policies. A founder can pressure-test positioning, outline an interview guide, or extract objections from pasted notes. A senior operator can ask for a decision memo structure before adding sensitive or proprietary material through an approved system. None of those jobs automatically needs Sol.

Free also includes more than a blank chat box. The live grid lists web search, Projects, shared projects, a built-in browser, plugins, study mode, limited vision, limited data analysis, limited file uploads, limited image generation, limited deep research, limited memory, limited Codex, and limited ChatGPT Work on desktop. The Free FAQ also confirms access to existing GPTs. The correct reading is “available with constraints,” not “equivalent to Plus.”

Winner: Free for text volume

Free wins plain text on price at every usage level. Dozens, hundreds, or thousands of ordinary Luna chats all carry the same $0 subscription fee. Plus cannot become cheaper on text volume because its fixed monthly fee starts at $20.

That does not make Free literally unmetered. OpenAI says unlimited text is subject to abuse guardrails. More importantly for ordinary buyers, tools sit behind separate meters. You can keep typing to Luna after reaching an image or upload limit, but the blocked tool remains blocked until its own reset.

The current pricing page lists unlimited everyday GPT-6 text on Free, subject to abuse guardrails. Some older paragraphs in the Free FAQ still name GPT-5.6 Luna; the newer release notes, Intelligent UI guide, and current pricing grid identify GPT-6 Luna for the rollout. Use the account’s model picker and in-product notices to confirm what has reached you.

Where Free still breaks

Free breaks first on tools, context, and availability rather than on basic typing.

The Instant context window is 27K, with an approximate maximum input of 12 pages of text. Context is the working material the model can consider during a task, not the long-term Memory feature that saves selected information across chats. Twelve pages can cover a memo, a short contract, or several interview transcripts. It is not enough for a large diligence pack, a long operating manual, or a multi-document analysis without splitting the material.

Free response time is also labeled as limited by bandwidth and availability. A casual user can wait. A manager preparing a board answer with a deadline minutes away cannot treat variable access as a harmless inconvenience.

The separate tool limits are deliberately dynamic. The current Free FAQ says data analysis, file and image uploads, image creation, and GPT access can each stop on their own schedule. OpenAI does not promise a stable public count for Free images or deep-research tasks on the current plan page. A fixed “five images” or “five research jobs” claim can become wrong without notice.

Who should keep Free

Keep Free when all of these are true:

• Most work is a normal text conversation rather than a tool-heavy workflow.
• Luna gives a usable answer after one or two clear revisions.
• The published 27K Instant context and approximate 12-page input estimate cover the normal source pack.
• Waiting for an image, upload, research, voice, or GPT meter does not threaten a deadline.
• You do not need hourly or exact-time schedules, event-triggered tasks, record mode, or Work from web and mobile.
• You are willing to check important claims rather than treating any model as a final authority.

Free is not the “personal” plan and Plus the “professional” plan. Free is the text-first plan. Plus is the reasoning-and-workbench plan. That distinction is more useful because it maps to an observable workflow instead of a job title.

ChatGPT Plus earns $20 on reasoning, not basic access

ChatGPT Plus is worth $20 when GPT-6 Sol or its surrounding workbench changes the quality, speed, or completeness of a valuable output. It is poor value when it merely makes an already-adequate Luna conversation feel more premium.

GPT-6 Sol powers the Plus Chat experience. In the model menu, GPT-6 replaces Latest; an explicit GPT-5.6 Sol selection stays on that older model. Plus includes Instant, Medium, and High, while Extra High and the Pro reasoning option require other eligible plans. Choose Medium or High when the task needs more reasoning: Instant on Plus does not automatically move up simply because you ask it to “think harder.” OpenAI’s model guide sets out those controls.

Winner: Plus for difficult reasoning

Plus wins when an error, omitted constraint, or shallow plan creates expensive rework. Sol supplies the paid Chat reasoning options. The practical test is whether those options improve a task with several interacting constraints, incomplete evidence, or a high cost of being wrong.

Consider three situations:

• A funded founder asks for ten homepage headlines. Luna is enough. The founder then asks which positioning claim survives a competitor map, customer-interview evidence, pricing constraints, and legal review. That second task is a Sol candidate because the hard part is reconciling evidence, not producing prose.
• A senior operator pastes a short meeting transcript and asks for action items. Free is enough. The operator then combines a policy, three spreadsheets, a prior decision log, and a deadline into a recommendation. Plus earns its fee through context, files, and deeper reasoning together.
• A solo technical builder wants a regular expression or a short explanation. Free can handle it. The builder wants a repository-level plan, iterative code work, and a finished handoff through Codex or Work. Plus provides the broader surface, though API usage remains separate.

The buying case needs evidence from the job you want completed. Use the same rubric you would use when comparing ChatGPT with another assistant: identical source material, a clear acceptance checklist, and the total time to a usable result. Count factual corrections, missed constraints, and review time on each plan.

A difficult task is a reason to try Plus, not proof that every Sol answer will beat Luna. If Free already produces a usable result after review, the paid model has not earned its fee on that task.

Plus buys a workbench around Sol

The model is only half the purchase. Plus adds or expands:

• A 54K Instant context window with about 40 pages of input.
• A 256K reasoning context window with about 320 pages of input.
• Files, data analysis, vision, deep research, and apps for deep research.
• Image generation with Thinking and interactive tables and charts.
• Expanded Memory and Memory with past chats.
• Up to 3 hours of GPT-Live-1 Voice per rolling 24 hours.
• Hourly and exact-time scheduled tasks, eligible event-triggered tasks, Sites where supported, record mode, and developer mode beta.
• Broader Codex and ChatGPT Work access, including eligible Codex Cloud workflows.
• Faster response service and priority during high-traffic periods.

That bundle explains why Plus can pay for itself even when Free text is unlimited. A recurring workflow rarely ends with the first answer. It starts with source material, moves through analysis, uses a tool, survives revisions, and finishes as something another person can review. Plus reduces the number of times that chain hits a plan boundary.

Plus still has limits

Plus is not a $20 unlimited-compute contract. The Plus FAQ says message caps and other limits can vary with system conditions, especially during high demand. OpenAI does not give a fixed Plus reasoning-message cap on that page. Deep research, Voice, images, and model access have their own allowance behavior. Plus is billed monthly, with no annual billing or multi-month prepayment option.

The limitation that matters most is less visible: Plus is still a personal workspace. It does not add an admin console, centralized member management, SAML SSO, a dedicated organizational workspace, or a business-data default. A CTO cannot turn five Plus subscriptions into a governed deployment by paying five separate invoices.

Plus wins a difficult individual workflow. It does not win every text prompt, every team purchase, or every data-governance decision.

The cost crossover: when Plus pays for itself

Plus crosses over only when better reasoning, larger context, faster access, or paid tools create at least $20 of monthly value. On raw text volume, Free remains cheaper forever.

That is unusual for a subscription comparison. The Free plan now offers unlimited ordinary Luna text, and Plus offers unlimited ordinary Sol text, both subject to guardrails. There is no message count where a $20 subscription undercuts a $0 subscription. The economic question is whether the paid workflow prevents enough time, rework, or missed output to cover the fee.

Cost per chat

Normalize the $20 fee across ordinary text usage:

• At 100 chats a month, Plus costs $0.20 per chat.
• At 250 chats, it costs $0.08 per chat.
• At 1,000 chats, it costs $0.02 per chat.

Free costs $0 per chat in subscription fees within the announced unlimited Luna access. The Plus numbers are amortizations, not prices OpenAI charges for individual messages. A month with 1,000 trivial prompts can still produce less value than a month with ten difficult decisions.

Cost per 1K output tokens

Ordinary Chat has no published per-token subscription tariff, so this utilization comparison needs an explicit assumption. Work and Codex have their own usage accounting. Assume 1,000 responses in a month and 500 output tokens per response. That produces 500,000 output tokens. Spread the $20 Plus fee across that output and the normalized cost is $0.04 per 1K output tokens. Free remains $0 per 1K output tokens in subscription fees.

This is not an API rate and should never be placed in an API budget. It is a way to compare subscription utilization. The moment the workflow uses an API, OpenAI bills that usage separately from Free or Plus.

Cost per image

OpenAI does not publish a stable current Free-versus-Plus image count on the plan grid. That prevents an honest guaranteed-capacity calculation. The only defensible subscription normalization is to state a workload and show the arithmetic:

• If the entire Plus fee is allocated to 20 images, the subscription works out to $1 per image.
• At 100 images, it works out to $0.20 per image.
• At 400 images, it works out to $0.05 per image.

Those numbers are not OpenAI image prices and do not guarantee that every account can generate those volumes. They also overstate image cost when the same subscriber uses Sol, research, Voice, files, or Work. Free remains $0 for every image completed within its changing allowance. Plus wins only when the extra capacity or paid image features matter more than the fee.

Cost per seat-month

One individual Plus account costs $20 a month or $240 a year. Five accounts cost $100 a month and $1,200 a year. Ten cost $200 a month and $2,400 a year. The matching Free subscription cost is $0.

This seat math exposes the wrong use of Plus. Five independent subscriptions do not become a managed team plan. If the buyer needs central controls, approved company context, or a no-training-by-default business environment, the decision has moved beyond Free versus Plus.

Time-saved break-even

The cleanest crossover is time. Divide $20 by the value of an hour:

• At $25 per hour, Plus must save 48 minutes per month.
• At $50 per hour, Plus must save 24 minutes per month.
• At $100 per hour, Plus must save 12 minutes per month.

A founder valued at $100 an hour does not need a dramatic productivity story. If Sol prevents one rewrite longer than 12 minutes, Plus has cleared the monthly threshold. A student valued at $25 an hour needs 48 minutes of genuine saving, not 48 minutes of enjoyable experimentation. A senior operator at $50 an hour needs 24 minutes, which can come from one file-analysis task that would otherwise be split and reassembled manually.

The crossover can also be measured in avoided failure. If Luna produces a usable result, Sol's incremental value is zero for that task. If Luna misses a constraint and the miss costs an hour of rework, Plus can repay several months at once. Track the whole outcome, not the model badge.

Three workload crossovers

A funded founder preparing a monthly investor update at a $100 hourly value needs Plus to remove 12 minutes. If Luna drafts the update cleanly from a short set of notes, Free wins. If Sol consistently reconciles the metrics, risks, prior promises, and current narrative with more than 12 fewer minutes of correction, Plus clears the threshold. The purchase is tied to one deliverable, so it can be checked again next month.

A senior operator valued at $50 an hour needs 24 minutes. Suppose the recurring job combines a policy document, a spreadsheet, and meeting notes. Free may handle each source separately, but splitting, prompting, and recombining the work can consume the saving. Plus wins when the larger context and file tools remove more than 24 minutes across the full process. If the operator still spends the same review time, the upgrade has not paid back.

A solo technical builder at $25 an hour needs 48 minutes. Free is the right plan for occasional explanations and small code questions. Plus becomes defensible when broader Codex access, Work, or Sol prevents almost an hour of monthly debugging or handoff work. It does not become defensible because the builder's application needs API calls; that remains a separate bill.

The thresholds scale with value, but the method stays constant: choose one recurring output, measure the before-and-after review time, and ignore usage that would have happened for curiosity rather than work. A high prompt count is not a business case. A repeated avoided failure is.

Models and context: Plus wins difficult, long work

Plus wins when the task is both hard and large. Free remains enough when the task is hard but compact, or large but easy to split cleanly.

The first difference is the model. Free gets GPT-6 Luna; Plus gets GPT-6 Sol, with a Thinking slider for its available reasoning levels. Pro, Business, and Enterprise also use Sol in Chat, while the eligible Pro reasoning option uses GPT-6 Astra. Plus does not include Pro reasoning in Chat: its Astra access is in Work and Codex. The model rows on the pricing grid span multiple products, so a checkmark is not permission to use every model in every tab.

Intelligent UI chooses a layout to fit the question: text, comparisons, diagrams, charts, or interactive controls such as a calculator. Answers can begin while the model continues thinking or using tools. It works from Instant through Extra High where the plan includes those levels, with no separate Intelligent UI quota; existing model and tool allowances still apply. It is unavailable with Pro reasoning, in Work, or in Voice. Plus users do not gain Extra High just because Intelligent UI supports it.

OpenAI still states that GPT-5.5 retires from ChatGPT, Work, and Codex on October 14, 2026; the API is unaffected. That is a separate retirement notice, not a claim that this rollout removes every older selectable model.

The second difference is context. Context is the working set a model can see during an answer. It is not the same as Memory, which stores selected information across chats. A larger context window lets a task carry more source material, instructions, conversation history, and intermediate work without discarding or compressing as much of it.

Winner: Plus for source-heavy analysis

The pricing grid publishes a 27K Instant context window and an approximate 12-page input maximum for Free. Plus doubles Instant context to 54K and about 40 pages. For reasoning, Free says the context varies, while Plus publishes 256K total context and an approximate 320-page input maximum.

The page counts are rough because formatting, tables, images, and token density vary. They are still useful buying markers.

A client brief around the 12-page Free estimate is a starting candidate for Instant. A board pack above that estimate but around the 40-page Plus estimate is a reason to compare paid capacity; Go also lists 54K Instant context, so context alone need not justify Plus. These are input estimates, not guaranteed document sizes. A long policy set may fit within Plus's approximate 320-page reasoning input, but fitting is not the same as understanding every line perfectly. Large-context work still needs source labels, targeted questions, and verification.

For a founder, the decision might look like this:

1. One customer interview and a short positioning note: Free.
2. Twelve interviews plus a competitive memo and pricing constraints: Plus.
3. A company-wide knowledge base with confidential material and access rules: neither personal plan; use an approved business environment.

For a solo technical builder:

1. A function, error message, or design pattern: Free.
2. A multi-file architecture review with a longer decision history: Plus.
3. A production application making API calls: separate API budget, regardless of the ChatGPT subscription.

Luna can still beat a badly framed Sol task

Paying for Sol does not repair a vague request. “Review this” gives the model no decision, audience, standard, or output shape. A clear Luna prompt can beat a careless Sol prompt because it gives the model a tractable job.

Give either plan the outcome, constraints, source hierarchy, and review criteria. Then judge whether Luna completes the job. If it does, keep Free. If it repeatedly loses constraints even with a clear request, the model-and-context upgrade becomes evidence-based.

Images, deep research, files, memory, and GPTs

Plus wins repeated tool use because Free makes the tools available but meters them more tightly and separately. Free wins exploration because it lets you discover which tools belong in a real workflow before paying.

Files and data analysis winner: Plus

Free includes limited file uploads, data analysis, and vision. Plus expands access, but the absence of a “limited” label is not an unlimited upload allowance. The file-upload documentation currently specifies 3 uploads a day for Free and a general ceiling of up to 80 files every 3 hours, which OpenAI can lower during peak periods. It does not display the remaining rolling upload quota. Projects allow up to 5 files on Free and 25 on Plus.

Size limits apply separately: documents and presentations can be up to 512 MB; text and document files also have a 2-million-token cap. Spreadsheets are approximately 50 MB and images 20 MB. These upload sizes are not promises that the model reads every token into its context window. Audio uploads require a paid plan. Data-analysis usage has no universal fixed Free-versus-Plus public job count on these pages.

A local-services owner uploading one price list and asking for clearer service descriptions can start on Free. A finance operator comparing monthly CSVs, reconciling exceptions, and producing a chart should expect Plus to be the practical plan. The benefit is not that Plus turns a weak spreadsheet into clean data. It is that the workflow can use the analysis surface more consistently and with more context.

Free tool meters can stop independently. Reaching an upload limit does not mean plain chat has ended. It means the file-dependent job is paused. The Free FAQ says upgrading resets the current Free usage rates, so an urgent buyer can clear the current Free meter by moving to a paid plan. That is useful in an emergency, but an emergency upgrade is poor procurement if the same blocker never returns.

Image generation winner: Plus for repeat production, Free for occasional use

Free offers limited image generation. Plus includes image generation, adds image generation with Thinking, and is positioned for more complex image work. The live plan grid does not promise a stable numeric Free-versus-Plus image allowance.

That missing number is a buying constraint. Do not plan a weekly image pipeline around a fixed count from an earlier plan version or one account's temporary UI. Use Free for occasional ideation or a small number of images. Use Plus when image work recurs and the $20 amortization makes sense at the account's observed capacity.

The craft boundary remains the same on both plans: generated visual output needs art direction, factual review, rights review where relevant, and a human decision about whether it is ship-ready. A higher plan can increase access. It does not make every image usable.

Deep research winner: Plus

Deep research is ChatGPT's multi-step research surface. It can search the web or specified sites, use uploaded files and enabled apps, propose a plan, and return a documented report with citations. Free has limited access. Plus has expanded access.

The current deep-research help page does not publish a universal fixed Free-versus-Plus task count. It says usage varies by plan, the remaining amount appears in the product, and a fixed monthly allowance can reset 30 days from first use. Treat the account counter as the source for procurement, not an allowance from a prior plan version.

Deep research earns Plus when research is an input to a decision, not when the user wants one quick fact. Search is faster for a current price, date, or definition. Deep research is appropriate for a market map, vendor diligence, policy comparison, or source-heavy briefing where the trail matters.

The user still owns verification. A cited report can cite the wrong passage, miss a source, or merge claims that belong to different scopes. Open the decisive sources and check them. Plus buys a stronger research workflow, not immunity from review.

Memory and context winner: Plus

Free includes limited Memory and limited Memory with past chats. Plus expands both. Memory is helpful for stable preferences, recurring context, and continuity. It does not retain every detail, and OpenAI publishes no fixed Free-versus-Plus memory capacity in the plan grid. It is not a substitute for a source of record.

A senior operator can let ChatGPT remember a preferred memo structure or recurring audience. The operator should not rely on Memory as the only home for a policy, financial figure, or approval decision. Put authoritative material in the appropriate system and bring it into the task when needed.

Scheduled work winner: Plus; new GPT creation is a tie

Free and Plus can discover and use existing GPTs, but neither personal plan can create or publish a new one. Editing an existing GPT depends on plan and permissions. The scheduled-tasks documentation allows eligible Free accounts 3 active tasks, run once or at most daily in flexible time windows. Plus allows 5 active tasks and adds hourly and exact-time schedules plus eligible event-triggered Work tasks. The pricing grid currently marks Free tasks unavailable, so confirm eligibility in the Scheduled page before depending on them. Plus also adds Sites where supported, record mode, developer mode beta, interactive tables and charts, and apps connecting to internal tools.

This is where Plus stops being “a smarter chat” and becomes a configurable individual workbench. A scheduled task can run a recurring prompt, while an event-triggered Work task can react to supported Gmail, Slack, or GitHub activity. Event-triggered tasks are capped at 30 runs an hour and 720 a day across the account’s event-triggered tasks; events may be grouped. Free and Go cannot create them. Those features earn money only when the job repeats. Paying for hourly automation for a task performed twice a year is configuration theater.

The governance caveat is unchanged. An existing GPT or scheduled task in a personal Plus account is not an approved enterprise agent. It lacks the admin and workspace controls that organizational use can require.

Voice, Work, Codex, and the retired Agent mode

Plus wins across Voice, Work, and Codex, but each surface has its own limits and device boundaries. Buying Plus without knowing which surface the workflow uses is an easy way to pay for the wrong capability.

Voice winner: Plus

The Voice guide gives Plus up to 3 hours of GPT-Live-1 Voice in Chat in each rolling 24-hour period; Free keeps limited GPT-Live-1 mini access with no fixed published hour count. The September 9 release notes retire the separate Instant, Medium, and High Voice intelligence levels. Plus no longer falls back to GPT-Live mini after 3 hours: the spoken session stops until time returns, though the thread can continue in text. This is Voice in Chat. Voice in Work on web and mobile draws from the Voice allowance as well as normal Work usage. Desktop Voice pricing instead charges connected Voice time against the Codex budget at $0.05 per minute, with task-model usage accounted for separately; Voice and tasks share the plan’s usage limits.

Voice is useful when typing is the bottleneck: rehearsing a meeting, interrogating an outline while walking, or thinking through a decision out loud. Plus wins when those sessions recur or need the stronger model options. Free remains enough for an occasional short conversation.

The video detail is easy to misunderstand. Live does not support video or screen sharing. Eligible subscribers can use video and screen sharing through Advanced Voice on iOS and Android. The plan grid lists Voice with video as unavailable on Free and available on Plus. Buying Plus for video means using the correct mobile Voice experience, not expecting every Voice surface to expose a camera. The current Voice guide does not publish a fixed Plus video or screen-sharing allowance.

Voice transcripts are also not verbatim records. Use dictation when the job is to capture editable text. Use Voice when the job is a live exchange.

ChatGPT Work winner: Plus

ChatGPT Work is the current agent for longer, multi-step work and finished deliverables. It can research, analyze information, and create documents, spreadsheets, presentations, reports, or Sites. Free gets limited Work access in the desktop app. Plus expands Work across desktop, web, and mobile.

That surface difference decides mobility. A solo consultant who wants to start a cloud Work task on the web, check it on mobile, and continue on desktop needs Plus. A Free user can explore limited Work from the desktop app but cannot assume the same cross-device cloud workflow.

Desktop and cloud also have different file boundaries. Work on web and mobile runs in the cloud. Desktop Work can use local files and desktop apps with permission when the plan allows. A task started on mobile cannot directly reach an arbitrary local file sitting on a laptop. The file needs to be available through the permitted task context.

Codex winner: Plus

Free has limited Codex access. Plus expands it and includes eligible Codex Cloud access. Codex remains the software-development surface for writing or debugging code, running tests and commands, reviewing changes, and working with repositories.

Codex remains a separate desktop view, but Codex Cloud can now start and continue tasks from desktop, web, and mobile. Cloud tasks can continue while your computer sleeps. Cloud is available to eligible Plus accounts and is not included with Free or Go. Supported desktop chats also remain reachable through mobile Remote; that is a different workflow from a task running in the cloud.

Work and Codex share an allowance. OpenAI publishes model-dependent estimates for local messages per five-hour period, explicitly not fixed message caps; cloud tasks may consume more and weekly limits may apply. Check the usage dashboard and pricing guidance before committing a workflow. Neither the Free/Plus grid nor those estimates guarantees a fixed number of completed Work or Codex tasks.

The subscription boundary matters again: ChatGPT Plus access to Codex is not an API allowance. A builder can use Codex through the plan and still receive a separate bill for an application that calls OpenAI's API.

Do not buy Plus for legacy Agent mode

OpenAI's current Agent help page says legacy ChatGPT agent mode is no longer available and directs users to ChatGPT Work for longer, multi-step tasks. The relevant Free-versus-Plus comparison is therefore limited desktop Work versus expanded Work on desktop, web, and mobile.

This is a named coverage gap in the product's transition. Older language about a fixed monthly Agent allowance no longer describes the current product boundary. Buy Plus for Work, hourly or event-triggered scheduled tasks, apps, Codex, and Sol. Do not buy it because an old feature list says “Agent mode” without checking what that now means.

The switching cost is workflow dependence

Moving from Free to Plus can immediately reset the current Free usage rates. The larger cost appears later, when a workflow depends on paid-only features: an hourly or event-triggered scheduled task, Sol reasoning, expanded context, Voice allowances, Work on web and mobile, or a recurring file and research process.

Before building around one of those features, name the output it supports and the fallback if the subscription ends. A simple prompt can move back to Luna. A monthly deliverable that depends on several paid surfaces has a higher switching cost. That can still be worth $20, but it should be visible.

Privacy is a tie, and that is the dealbreaker

Free and Plus have the same core personal-plan privacy boundary, so Plus does not win sensitive organizational work. Both list a training opt-out. Neither includes SAML SSO, unified billing, a dedicated workspace, an admin console, bulk member management, or admin roles.

The data-controls documentation says turning off Improve the model for everyone applies across the account. Chats can remain in history while not being used to improve ChatGPT. An unsaved, non-personalized Temporary Chat does not appear in history, does not create memories, and is not used to improve models while it remains temporary; OpenAI may keep a safety copy for up to 30 days. Saving it converts it into a regular chat that follows the account's personalization and model-improvement settings.

Those controls are useful for individual use. They do not create an organizational policy, enforce who can connect which source, or provide a central place to manage departure, access, retention, and approved data classes.

For a mid-market CTO, the decision is straightforward. If employees will use confidential company context, regulated data, customer records, unreleased financial information, or internal systems, compare business workspaces and the organization's requirements. Do not treat Plus as “Business Lite” because it costs money.

For an individual professional, set the training preference deliberately and use Temporary Chat when its behavior fits the task. Do not paste passwords, API keys, recovery codes, or material the organization has not approved for a personal AI workspace. A model subscription is not permission to move data.

Privacy is a tie that can outweigh every capability difference. If governance is the deciding axis, leave this comparison and choose an organizational plan that satisfies the requirement.

How to upgrade without paying on instinct

Do not upgrade until you can name the blocked output, measure the cost of the block, and identify the Plus feature that removes it. A one-month decision cycle is enough for most individual buyers.

1. 1

   Start with the post-rollout Free plan

   Use Luna for the real recurring job. Keep the source pack, instructions, and expected output consistent. Do not compare a polished Sol prompt with a careless Luna prompt.

2. 2

   Record the blocker

   Classify every failure as model quality, context, a separate tool meter, response availability, or a missing feature. “Plus feels better” is not a blocker. “The pack exceeds Free's approximate 12-page Instant input” is.

3. 3

   Price the full interruption

   Count prompting, waiting, splitting files, reassembly, correction, and review. Compare the monthly total with 48 minutes at $25 an hour, 24 minutes at $50, or 12 minutes at $100.

4. 4

   Buy one month and retest

   Use the same task on Sol and the paid workbench. Keep Plus if it removes more than $20 of monthly friction or improves a high-value outcome enough to justify the fee. Return to Free if the result remains equivalent after review.

Who should not switch

Do not switch to Plus when plain Luna text completes the job, tool caps are rare, and waiting has no cost. The current unlimited-text allowance keeps the decision focused on the job rather than a prompt count.

Do not switch solely for a larger model name. A clean Luna workflow can outperform a confused Sol workflow. The paid model should be solving a measured quality problem.

Do not switch a team to a pile of personal Plus accounts when the actual need is governance. Five Plus accounts cost $100 a month without becoming a centrally managed workspace.

Do not switch for a fixed image or research count copied from an old source. OpenAI's current public pages use “limited” and “expanded” language and direct users to in-product counters. Observe the account's real allowance before building a commitment around it.

Do not switch because of legacy Agent mode. The current long-task product is Work, and its device access differs by plan.

Who should switch now

Switch when one of these is already true:

• Sol prevents repeated reasoning failures in a valuable task.
• The 54K Instant or 256K reasoning context avoids recurring document splitting.
• Files, data analysis, deep research, images, or Voice hit Free meters often enough to interrupt work.
• An hourly or event-triggered scheduled task supports a genuine recurring outcome.
• ChatGPT Work on web and mobile is necessary rather than convenient.
• Faster response access matters during deadline-bearing work.

The strongest buying signal is repeated, measured friction. The weakest is fear of missing a feature.

Frequently asked questions

What is ChatGPT Plus?

ChatGPT Plus is OpenAI’s $20 monthly subscription with broader model and tool access, faster responses, and priority during high traffic. The GPT-6 Chat rollout uses Sol on Plus. API usage is billed separately. Sources: Plus FAQ and Intelligent UI guide.

How do rate limits work on the Free tier?

Everyday text is unlimited subject to abuse guardrails. Uploads, image generation, Voice, data analysis, GPT access, and other tools keep separate limits. OpenAI does not publish a fixed count for every tool; the product shows applicable limit notices and reset timing. The upload guide does publish 3 Free uploads per day. Sources: Free FAQ and upload limits.

What Voice options are available?

Live uses GPT-Live-1 mini on Free with a changing, unpublished hour limit. Plus includes up to 3 hours of GPT-Live-1 per rolling 24 hours; the release notes say Plus no longer falls back to mini at its limit. Eligible subscribers use Advanced on iOS or Android for video and screen sharing, which Live does not support. Sources: Voice guide and release notes.

How is ChatGPT Go different from Plus?

Go costs $8 a month in the US; Plus costs $20. Both Free and Go use GPT-6 Luna in the Chat rollout, while Plus uses GPT-6 Sol. Go expands uploads, images, data analysis, and memory beyond Free. The Voice guide lists 3 hours of GPT-Live-1 mini for Go versus GPT-Live-1 for Plus per rolling 24 hours. Sources: pricing, Go FAQ, Intelligent UI guide, and Voice limits.

Is ChatGPT agent mode available on the Free plan?

For legacy ChatGPT agent mode, OpenAI’s current lead notice says the product is no longer available and directs users to Work. Older Agent limits elsewhere on that page are not Work allowances. Free has limited desktop Work; Plus expands Work to desktop, web, and mobile. Sources: Agent help and plan grid.

Can I create a GPT as a Free tier user?

No. New GPT creation and publication are unavailable on personal Free, Go, Plus, and Pro accounts. Existing GPTs can still be used; editing depends on plan and permissions. Source: Free FAQ.

How does Intelligent UI work across reasoning settings?

It works from Instant through Extra High where the account’s plan includes those levels. Plus includes Medium and High, not Extra High or Pro reasoning in Chat. Pro reasoning uses GPT-6 Astra and does not support Intelligent UI. Intelligent UI has no separate quota; normal plan and tool limits apply. Sources: Intelligent UI guide and model guide.

Want a faster way to match each AI subscription to a recurring business outcome? Get the AI Tools Map for Business Owners.