The message caps are real. But the ceiling that costs you most is not a message count.
ChatGPT Plus is $20 a month, monthly only, and OpenAI publishes almost no usage numbers for it. It does publish exact numbers for the plans above it — and that is how you work out what Plus is actually missing. The binding constraint on Plus is rarely how many messages you send. It is which reasoning level and which models you are not allowed to reach at all.
ChatGPT Plus at $20/month includes GPT-5.6 with Medium and High reasoning, expanded messages and uploads, expanded deep research, and full Codex access. It does not include Extra High reasoning or any Pro model — those begin at Pro $100. Since August 25, 2026, Codex and ChatGPT Work on Plus run against a five-hour window as well as a weekly cap. OpenAI publishes no message count for Plus, and states plainly that there is no setting to bypass an allowance and that support cannot reset one.
TL;DR — ChatGPT Plus at $20/month buys GPT-5.6 with Medium and High reasoning, full Codex access, and expanded uploads and deep research — but not Extra High reasoning and not the Pro models, which start at $100. OpenAI never publishes a Plus message count, though it publishes precise weekly caps for Pro, which is how you can infer what you are actually short of. Since August 25, 2026 a five-hour window applies to Codex and Work on Plus again. Free to read, no affiliate relationships. Verified September 3, 2026.
Updated October 2026. Maintained by a small team of AI super-users who teach multi-AI research and study workflows to researchers, students, and professionals — no affiliate relationships. About this guide →
What this pattern tells you: the price of Plus has not moved, but the plan above it, the plan below it, and the shape of its own limits all have. If you are trying to decide whether Plus is enough, the honest starting point is that “enough” is a moving target that OpenAI adjusts without changing your invoice.
OpenAI is specific about model access and deliberately vague about volume. Both halves are useful once you stop expecting the second one to exist.
| Fact | What OpenAI states |
|---|---|
| Price | $20/month, billed monthly. There is no annual option. |
| Reasoning access | Medium and High reasoning included. Extra High is not included on Plus. |
| Pro models | Not included on Plus. GPT-6 Pro and GPT-5.6 Sol Pro begin at Pro $100. |
| Codex | Full Codex access, sharing its allowance with ChatGPT Work, ChatGPT for Excel, and Workspace Agents. |
| Other inclusions | Expanded messages and uploads, expanded deep research, image creation, voice, memory, priority access at busy times. |
| Message caps | “Plus subscriptions may include usage limits such as message caps, especially during high demand. These limits may vary based on system conditions.” No figure is given. |
| At the cap | “ChatGPT may continue with another available reasoning model” rather than stopping outright. |
| Overrides | “There is no setting to increase or bypass a model’s usage allowance,” and OpenAI Support does not reset ChatGPT or Codex limits. |
| Where to check | Settings, or the Codex usage dashboard at chatgpt.com/codex/settings/usage. |
Official facts: What is ChatGPT Plus? · GPT-5.6 and GPT-6 Pro in ChatGPT · Using Codex with your ChatGPT plan.
Here is the part most limit guides skip, because it is not a number about Plus — it is a boundary around it. Two things on ChatGPT are not rationed on Plus. They are absent.
OpenAI documents four reasoning levels. Plus reaches Medium and High. Extra High is listed as included on Pro and Business, and not on Plus. No amount of restraint elsewhere in your week unlocks it, and no credit purchase adds it.
GPT-6 Pro and GPT-5.6 Sol Pro are Pro-tier models. On Plus they are not in the picker. This is the single clearest reason to move up a tier, and the only one that a heavier Plus subscription can never solve.
The useful trick is that OpenAI publishes exact Pro allowances even while publishing none for Plus. That turns the paid ladder into a measuring stick: you can read precisely what the tiers above buy, and infer the shape of what you are missing.
| Plan | Price | Reasoning | Published Pro-model allowance |
|---|---|---|---|
| Free / Go | $0 / low | Think uses GPT-5.6 Luna | None. Unlimited everyday text chats, subject to abuse-prevention safeguards. |
| Plus | $20 | Medium + High | None. No Pro models at any volume. |
| Pro $100 | $100 | Medium, High, Extra High | 50 messages per week, shared across GPT-6 Pro and GPT-5.6 Sol Pro. |
| Pro $200 | $200 | Medium, High, Extra High | 200/week GPT-6 Pro; 170/day GPT-5.6 Sol Pro; 200/day combined. |
| Business Standard | Per seat | Medium, High, Extra High | 15 Pro messages per month, shared. |
| Business Premium | Per seat | Medium, High, Extra High | 50 Pro messages per week, shared. |
Official facts: OpenAI — GPT-5.6 and GPT-6 Pro in ChatGPT · OpenAI — About ChatGPT Pro tiers.
For Codex and ChatGPT Work on Plus, two ceilings run at once. They fail differently, and confusing them is the source of nearly every angry forum thread on the subject.
A rolling budget that refills on its own schedule. It is what stops you mid-task, and it is deliberately smaller than your weekly share — OpenAI reinstated it precisely so a single afternoon could not consume the week.
The larger allowance the five-hour window is rationing on your behalf. You can be blocked by the window while most of the week is still untouched, which feels like a bug and is the design.
That is a Plus subscriber writing on OpenAI’s own developer forum in 2026, describing a single prompt that consumed an entire five-hour allowance while more than 80% of the weekly allowance sat unused. The task did not finish.
Another wrote that he “barely used it for 1.5 hours coding my personal project and I ran out of 5 hour limit after just 3 prompts.” A third asked OpenAI directly why a limit is displayed in hours when what is actually being spent is tokens — repository size, context, model choice, task complexity. An OpenAI moderator acknowledged the confusion and described the mechanism accurately: a five-hour time window, and a number of tokens that can be spent within it.
These complaints are not evidence that the plan is bad. They are evidence that a limit labeled in hours will be read as an allowance of hours, by nearly everyone, until they hit it.
For several weeks in mid-2026, Plus ran on a weekly cap alone and the five-hour window was gone. Then it returned. The stated reasoning is worth reading closely, because it explains a design most people experience as arbitrary.
OpenAI’s engineering lead gave two reasons. The first is infrastructural: the window “allows us to smoothen the load on our compute, allowing to keep the plan generous in terms of weekly usage.” The second is about users: Plus subscribers, described as relatively casual and new, sometimes “accidentally eat through their whole weeks usage and then are confused.”
Take that at face value and it is a real trade. Without the window, a single heavy Monday can cost you Thursday and Friday. With it, you get interrupted more often but stranded less often. Whether that is the right trade depends entirely on whether your work arrives in bursts or spreads across the week — which is something only your own measurement can tell you.
If your work is bursty and the five-hour window is what keeps stopping you, the fix is not a bigger plan. It is splitting the burst, or moving part of it to a tool with a different reset shape.
Reported by 9to5Mac Developers Digest — Codex plan tiers and Pro $500 (reported, September 29) Morph — Codex pricing tracker (reported, updated October 7), quoting OpenAI engineering lead Tibo Sottiaux.
OpenAI names the variables even where it withholds the totals. Every factor below comes from its own documentation; the ordering is ours.
| Factor | Why it costs | What to do instead |
|---|---|---|
| Where the task runs | Codex distinguishes local from cloud execution, and they do not cost the same. | Check the usage dashboard after each mode before assuming they are interchangeable. |
| Context size | A large repository or long thread is re-read on each task. Loading context spends allowance before any work happens. | Scope the task to the files it needs. Do not point an agent at the whole repo out of convenience. |
| Reasoning level | Higher reasoning spends more per response, whether or not the task required it. | Reserve High for problems that genuinely branch. Most editing and drafting does not. |
| Task complexity | An agent that plans, edits, runs and retries spends far more than one that answers. | Break large agent tasks into checkpoints you can inspect between steps. |
| Tools | Deep research and tool calls run several internal steps per visible answer. | Use deep research for surveys, not for facts you could look up in one search. |
| Shared surfaces | Codex, ChatGPT Work, ChatGPT for Excel and Workspace Agents draw on the same allowance. | Remember that a spreadsheet session quietly reduces what is left for coding this week. |
| Uploads and images | Uploads and image generation are metered separately from text and remain limited on every tier. | Batch image work rather than iterating one small change at a time. |
Official facts: OpenAI — Using Codex with your ChatGPT plan.
Credits are OpenAI’s overflow valve for personal plans. They are genuinely useful and they are not a way to buy your way past the ceiling described above.
| Question | Answer |
|---|---|
| Order of use | “Your plan’s included usage is used first. After you hit plan limits, usage draws from your credit balance.” |
| Who can buy | Plus and Pro, plus a limited group of Free and Go users. |
| What they cover | Codex, ChatGPT Work, and ChatGPT for Excel — supported features, after included limits are exhausted. |
| What they do not cover | Capability. Credits do not unlock Extra High reasoning or the Pro models on a Plus plan. |
| Typical consumption | OpenAI cites roughly 5–20 credits per message for typical tasks. |
| Expiry | “Credits are valid for 12 months from purchase. Unused credits expire and do not roll over.” |
| Automation | Auto-reload can maintain a minimum balance, bounded by a monthly spending cap you set. |
Official facts: OpenAI — Using credits for flexible usage OpenAI — Paid weekly Work and Codex rate limit resets OpenAI — ChatGPT release notes (GPT-6 Astra rollout).
Because OpenAI publishes no Plus figure and explicitly says limits vary with system conditions, the only number worth having is one you produced yourself. Two minutes a day for five working days.
Run the same protocol on Claude Pro using Settings → Usage and the two results become directly comparable — which is the only sound basis for deciding whether a second $20 subscription earns its place.
This is why the protocol above matters more than any table. A figure from your own dashboard over five days is calibrated to your work; a figure from a blog post is calibrated to someone else’s.
These are the questions people actually type when a limit stops them mid-task.
OpenAI does not publish a message count for Plus. Its documentation says Plus may include message caps, especially during high demand, and that these limits may vary based on system conditions. It directs users to the model picker and their own usage dashboard rather than to a number, so any specific figure you find online is an estimate.
No. GPT-6 Pro and GPT-5.6 Sol Pro are Pro-tier models and are not available on Plus at any volume. The lowest tier that includes them is Pro at 100 dollars a month, which publishes an allowance of 50 messages per week shared between the two.
Medium and High. Extra High reasoning is documented as included on Pro and Business plans and not on Plus. This is a capability boundary rather than a usage cap, so it cannot be lifted by buying credits or by using less of anything else.
Because the window is a token budget with a time-based reset, not an allowance of hours. Consumption depends on the model, where the task runs, task complexity, context size, reasoning level, and tools. A single agent task against a large repository can spend more than an afternoon of short questions.
Yes, on August 25, 2026, for Codex and ChatGPT Work, after several weeks of a weekly-only cap. OpenAI said the window smooths load on its compute and keeps weekly usage generous, and that Plus users otherwise sometimes consume a whole week's allowance by accident.
OpenAI Support does not manually reset or raise usage limits. However, eligible Plus and Pro personal accounts can now buy an instant weekly Work/Codex reset. A purchased reset immediately restores both five-hour and weekly usage, pulls the normal weekly allowance forward, and starts a new weekly period with the next Work or Codex request. Credits remain a separate pay-as-you-go option, and higher plans add more included capacity.
Yes, if the option is available on your account. Eligible Plus and Pro personal accounts can purchase an instant reset from Usage settings or, after reaching the weekly cap, from an in-app offer. The reset applies immediately, cannot be banked for later, and changes the date of your next automatic weekly reset.
ChatGPT may continue with another available reasoning model rather than stopping outright. This is worth knowing because it can look like a quality drop rather than a limit, and it is easy to misread as the model getting worse.
Yes. Credits are valid for 12 months from purchase, and unused credits expire without rolling over. If you enable auto-reload, it is possible to accumulate a balance you never spend, so it is worth checking before the twelve-month mark.
Yes. Codex shares usage with ChatGPT Work, ChatGPT for Excel, and Workspace Agents. A heavy spreadsheet session quietly reduces what is left for coding in the same period, which is a common source of unexplained limits.
No. Plus is 20 dollars a month billed monthly, and OpenAI does not currently offer an annual option for Plus or for either Pro tier.
It depends on which ceiling binds you. If you need volume, Plus is generous and credits cover spikes. If you need Extra High reasoning or a Pro model, Plus cannot supply either at any usage level, and the honest answer is that the 20 dollar plan is not the product you need.
Open Settings, or the Codex usage dashboard at chatgpt.com/codex/settings/usage. Reading it at the start and end of each block of work for a week gives you a burn rate per activity, which is more useful than any published estimate.
Product facts were checked against first-party OpenAI documentation on September 3, 2026. Community links are included as qualitative evidence of how limits are experienced, not as specifications. Where a figure is an estimate rather than a published fact, the sentence using it says so.
The cheaper one is usually routing: giving each system the work it is actually best at, with durable handoffs, shared state, and a verification role that is not the same model that wrote the draft. That is what Multi-AI Systems is for.
When a limit pushes work to a different tool, the data goes with it. A more available model is not automatically the right destination for confidential, regulated or client-sensitive material. Check data handling and organizational policy before moving project state between providers.