ChatGPT Plus vs Pro
ChatGPT Plus is $20 a month, Pro starts at $100. We checked OpenAI's current limits and ran 10 business tasks through the Plus and Pro models side by side.
ChatGPT Plus costs $20 a month. ChatGPT Pro starts at $100 a month (5x the Plus usage allowance and 50 GPT-6 Pro messages a week) and goes to $200 a month (20x the allowance and 200 GPT-6 Pro messages a week), although OpenAI paused new $200 sign-ups on 10 September 2026. Pro is worth it if you hit Plus limits every day or need GPT-6 Pro inside Chat. On ten everyday business tasks we ran side by side, the Pro-only models did not produce better results than the model Plus already includes.
Key Takeaways
- Plus is $20 a month. Pro is $100 or $200 a month. Both Pro tiers have the same features; the difference is how much you can use them.
- Pro $100 gives 5x the Plus usage and 50 GPT-6 Pro messages a week. Pro $200 gives 20x the Plus usage and 200 GPT-6 Pro messages a week.
- Plus does not include GPT-6 Pro in Chat. It includes GPT-6 Astra and GPT-5.6, and GPT-6 Astra inside ChatGPT Work and Codex.
- New Pro $200 subscriptions and upgrades are paused as of 10 September 2026. Existing $200 subscriptions keep renewing. Pro $100 is still open.
- We ran 10 business tasks through GPT-5.6 Sol (the Plus-level model), GPT-5.6 Sol Pro and GPT-6 Astra Pro (the Pro-only models) with identical prompts. 27 of 30 outputs passed, 3 were partial, none failed and none contained an invented fact. The three models gave the same answers on every calculation and formula.
- The Pro-only models cost 4.2x and 12.9x more credits than GPT-5.6 Sol for the same ten tasks, with no quality gap we could measure on this type of work.
- Pick Pro if you are a heavy daily user who hits Plus limits, runs long Codex or deep research jobs, or wants GPT-6 Pro on hard reasoning problems. Pick Plus for writing, analysis, spreadsheets and documents.
ChatGPT Plus vs Pro at a glance
All figures below come from OpenAI's own pages, checked on 17 September 2026. OpenAI changes plans often, so treat the links as the source of truth if you are reading this later.
| Plan | Price | Usage vs Plus | GPT-6 Pro in Chat | Who it fits |
|---|---|---|---|---|
| ChatGPT Plus | $20 / month | Baseline | Not included (GPT-6 Astra in ChatGPT Work and Codex) | Most individual users and small teams doing writing, analysis and documents |
| ChatGPT Pro $100 | $100 / month | 5x Plus | 50 messages per week, shared with GPT-5.6 Sol Pro | Heavy daily users who hit Plus limits and want some GPT-6 Pro access |
| ChatGPT Pro $200 | $200 / month (new sign-ups paused since 10 Sep 2026) | 20x Plus | 200 messages per week, plus a separate 170 GPT-5.6 Sol Pro messages per day (200 per day combined) | Power users running Codex, deep research and GPT-6 Pro all day |
Sources: What is ChatGPT Plus, What is ChatGPT Pro, GPT-5.6 and GPT-6 Pro in ChatGPT and chatgpt.com/pricing.
What you actually get for the extra money
OpenAI is explicit that both Pro tiers include the same core capabilities and that the main difference is usage allowance. In practice Pro buys you four things Plus does not have:
- GPT-6 Pro in Chat. This is the reasoning setting powered by GPT-6 Astra. Pro $100 gets 50 messages a week, shared with GPT-5.6 Sol Pro. Pro $200 gets 200 messages a week for GPT-6 Pro and a separate 170 a day for GPT-5.6 Sol Pro, capped at 200 a day across both.
- Higher allowances everywhere else. 5x or 20x the Plus usage across the models Plus also has, subject to what OpenAI calls guardrails against abuse.
- Maximum Codex tasks and deep research. The pricing page lists both at the maximum tier for Pro. ChatGPT Work and Codex have separate allowances from Chat.
- Headroom. Plus usage limits vary with system load. If you have been rate limited on Plus mid-task, that is the problem Pro solves.
Two things Pro does not give you: a different feature set (the help centre says both tiers share the same capabilities), and a guarantee of better answers on ordinary work. That second point is what we tested.
The Pro $200 pause, explained
As of 10 September 2026, OpenAI is not taking new sign-ups or upgrades to the Pro $200 plan. Existing $200 subscriptions continue to renew, and the $100 tier remains open. OpenAI has not published a date for reopening. If you were planning to buy the $200 plan for the 200 weekly GPT-6 Pro messages, your options today are Pro $100 (50 a week) or a Business plan (Business Premium also lists 50 a week). Source: What is ChatGPT Pro.
How we tested Plus-level and Pro-level models on real work
Comparing subscriptions by reading feature lists tells you what you are allowed to do, not what you get back. So we ran the three relevant OpenAI models against each other on the same ten business tasks in the Krater Model Arena, which sends one prompt to several models at once and shows the answers side by side.
Tested: September 2026. Round run on 17 September 2026. Results are for the model versions named below and will be rerun when OpenAI ships new versions.
- Models:
openai/gpt-5.6-sol(the GPT-5.6 model a Plus subscriber gets),openai/gpt-5.6-sol-proandopenai/gpt-6-astra-pro(the two models behind the Pro tiers' weekly allowances). See the GPT-5.6 Sol Pro and GPT-6 Astra Pro model pages. - Settings: Manual Battle mode, default temperature, no system prompt, no attachments, no web browsing, identical prompt text for all three models.
- Date: 17 September 2026, 13:34 to 14:03 UTC, from one Krater account.
- Measurements: credits charged and elapsed time were read from the saved comparison records, not from a stopwatch. Each comparison runs the three models at the same time, so the elapsed time is the time until the slowest of the three finished.
- Scoring: every output was checked against an answer key (for the calculation and formula tasks) or the instruction list in the prompt (for the writing tasks). Pass means every requirement met and every fact correct. Partial means a requirement missed or a fact incomplete. Fail means a wrong number, a broken formula or an invented fact.
One important caveat: we compared the models, not the ChatGPT app. ChatGPT Pro also gives you higher usage limits, Codex and deep research allowances, and OpenAI's own routing and tools around these models. Our test answers the narrower question of whether the Pro-only models give better answers than the Plus-level model on typical business work. Full details on how we run and score comparisons are on our methodology page.
Results: 10 tasks, 30 outputs
| # | Task | GPT-5.6 Sol (Plus) | GPT-5.6 Sol Pro (Pro) | GPT-6 Astra Pro (Pro) | Credits | Time |
|---|---|---|---|---|---|---|
| 1 | Cold email, bookkeeping firm to trucking company, under 150 words | Pass (92 words) | Pass (91 words) | Pass (106 words) | 11 | 11.7 s |
| 2 | Gross margin table for six months, name the worst month | Pass, all 12 figures correct | Pass, all 12 figures correct | Pass, all 12 figures correct | 17 | 14.2 s |
| 3 | Refund policy into 5 plain-English bullets, meaning unchanged | Pass | Pass | Pass | 8 | 10.6 s |
| 4 | One-page website redesign proposal outline with three tiers | Pass | Pass | Pass | 27 | 24.9 s |
| 5 | Current Plus and Pro prices plus the weekly GPT-6 Pro allowance, with sources | Partial | Partial | Partial | 21 | 23.2 s |
| 6 | Excel formula counting unique customers in a date range | Pass | Pass | Pass | 19 | 16.7 s |
| 7 | Part-time UK support job description, maximum 250 words | Pass (230 words) | Pass (227 words) | Pass (233 words) | 15 | 16.0 s |
| 8 | Rewrite a legal maintenance notice, keep every fact | Pass | Pass | Pass | 7 | 7.7 s |
| 9 | Five-step chargeback SOP with owner, deadline and evidence per step | Pass | Pass | Pass | 29 | 30.1 s |
| 10 | Reprice after a 12 percent cost increase, two margin rules | Pass, all 3 figures correct | Pass, all 3 figures correct | Pass, all 3 figures correct | 17 | 14.3 s |
| Totals | All 10 tasks | 9 pass, 1 partial | 9 pass, 1 partial | 9 pass, 1 partial | 171 | 169.4 s |
Credits and time are per comparison (all three models together). Invented facts across 30 outputs: 0. Wrong numbers: 0. Broken formulas: 0.
The numbers were identical
On the two calculation tasks all three models produced the same correct figures. June was the worst margin month at 29.55 percent in task 2, and in task 10 all three arrived at a new cost of $29.12, a $44.80 price to keep a 35 percent margin, and $43.12 to keep the original $14 dollar margin. On the Excel task all three returned the same formula, character for character:
=IFERROR(ROWS(UNIQUE(FILTER(A2:A100000,(A2:A100000<>"")*(B2:B100000>=DATE(2026,7,1))*(B2:B100000<DATE(2026,10,1))))),0)All three also flagged the Microsoft 365 or Excel 2021 requirement for dynamic arrays, and all three used the "before 1 October" trick so that order dates with timestamps on 30 September are counted.
The writing was close to interchangeable
The cold emails, refund bullets, job descriptions and maintenance notice from GPT-5.6 Sol and GPT-5.6 Sol Pro often shared whole sentences. Both job descriptions opened with "We're an online furniture store dedicated to helping customers create homes they love." GPT-6 Astra Pro wrote its own phrasing more often, and its refund bullets stayed closest to the original policy by keeping Final Sale items and gift cards as two separate points, where the other two merged them into one bullet. That is a style difference, not a correctness one: the meaning survived in all three.
Where all three fell short: task 5
We asked each model for today's Plus and Pro prices and the weekly GPT-6 Pro allowance on the $200 plan, with a source for each number, and told them to say so if unsure. All three gave $20 for Plus (correct) and $200 for Pro. That is one of the two Pro prices, but none mentioned the $100 tier, which is why we scored all three partial. All three then declined to state the weekly allowance because they could not verify it. That is the right behaviour without web access, and it is why none of these outputs counts as a hallucination. The verified answer is 200 messages a week on Pro $200 and 50 on Pro $100, from the OpenAI help centre.
What the Pro-level models cost for the same answers
Because every comparison ran all three models on the same prompt, we can split the bill. Krater records the underlying usage cost of each model per response and converts it to credits.
| Model | Credits for all 10 tasks | Relative to GPT-5.6 Sol | Score |
|---|---|---|---|
| GPT-5.6 Sol (Plus level) | 9.17 | 1x | 9 pass, 1 partial |
| GPT-5.6 Sol Pro (Pro only) | 38.71 | 4.2x | 9 pass, 1 partial |
| GPT-6 Astra Pro (Pro only) | 118.28 | 12.9x | 9 pass, 1 partial |
The round was billed at 171 credits in total because each comparison is rounded up to a whole credit. The ratio is the point: on this kind of work the Pro-only models cost 4x to 13x more per answer and did not score higher. That mirrors what OpenAI itself says about its tiers. You are paying for allowance and for headroom on hard problems, not for better cold emails.
On time, the whole round took 169.4 seconds of waiting across ten comparisons, an average of 16.9 seconds each, with the quickest (the rewrite) at 7.7 seconds and the slowest (the five-step SOP) at 30.1 seconds. Since the three models ran at the same time, each figure is the wait for the slowest model on that task.
ChatGPT allowances vs Krater credits
ChatGPT sells you a monthly seat with usage caps: a number of messages per week for GPT-6 Pro, a multiplier on general usage, and separate pools for Codex and Work. You do not see what any single message costs. Krater works the other way round. Every Krater plan comes with a monthly credit balance (Pro has 1,500, Ultra 4,000 and Max 10,000) and each response is charged in credits according to what the model actually used, so a short rewrite on GPT-5.6 Sol costs a fraction of a credit while a long GPT-6 Astra Pro SOP costs around 20. The same 400+ models sit next to each other in one chat, so you choose the heavier model for the task that needs it instead of paying for the allowance all month. The Model Arena is the practical tool for that decision: run your own prompts through the three models above and see whether the Pro-level answer is different enough to matter for your work.
Who should pick ChatGPT Pro
- You hit Plus usage limits most days and the interruption costs you real time.
- You run long Codex or deep research jobs and need the maximum allowance for those.
- You work on hard multi-step reasoning (complex analysis, research synthesis, difficult code) where a stronger reasoning setting has shown you a difference on your own tasks.
- You want GPT-6 Pro inside Chat rather than through Codex or Work.
Who should stay on ChatGPT Plus
- Your work is emails, documents, summaries, spreadsheets, policies and proposals. On ten tasks of that type we found no quality gap.
- You rarely see a usage limit on Plus.
- You would rather spend the difference on a second tool or on the people using it.
When neither is the right tool
- Live prices and allowances. None of the three models could give the weekly GPT-6 Pro allowance without web access, and all three missed the $100 tier. For anything that changes weekly, open the vendor page or use a model with browsing switched on.
- Spreadsheet work on real data. The formula was correct, but the models could not see your workbook. Paste the formula into Excel and check it against a filter on a known month before trusting it.
- Dispute deadlines and legal notices. The chargeback SOPs were sensible, but the deadline that counts is the one on your processor's notice. Use the output as a template and fill in your processor's actual windows.
The 10 prompts we used
Copy these into your own tool to reproduce the test. They are the exact text we sent, with the figures we used.
- Write a cold email from Northline Books, a small bookkeeping firm, to the owner of a regional trucking company with 12 trucks. Goal: a 20 minute call. Keep it under 150 words, plain language, no buzzwords, include a subject line and one specific reason a trucking company benefits from outsourced bookkeeping.
- Here is monthly data for a small online store. Revenue / cost of goods: Jan 18,400 / 11,200; Feb 16,900 / 10,900; Mar 21,300 / 12,800; Apr 19,750 / 13,100; May 23,100 / 13,900; Jun 17,600 / 12,400. Calculate gross margin in dollars and percent for each month, name the worst month by margin percent, and give one sentence on what to look at first. Show the numbers in a table.
- Turn this refund policy into 5 short bullets for a help centre page, plain English, nothing added or removed in meaning: 'Items may be returned within 30 days of delivery provided they are unused and in original packaging. Refunds are issued to the original payment method within 10 business days of us receiving the item. Return shipping is paid by the customer except where the item arrived damaged or we sent the wrong item, in which case we provide a prepaid label. Sale items marked Final Sale cannot be returned. Gift cards are non refundable.'
- Create a one page proposal outline for a website redesign for a dental clinic with 3 locations. Include: goals, scope, three package tiers (Essential, Growth, Complete) with what each contains but no prices, timeline, assumptions, and what we need from the client. Use headings and short bullets.
- What do ChatGPT Plus and ChatGPT Pro cost per month today, and how many GPT-6 Pro messages per week does the Pro $200 plan include in Chat? Give the source URL for each number. If you are not sure, say so instead of guessing.
- In Excel, column A has customer email addresses and column B has order dates. Write one formula that counts how many unique customers placed at least one order between 1 July 2026 and 30 September 2026 inclusive. Explain how it works in two sentences and mention any Excel version requirement.
- Write a job description for a part time customer support representative at an online furniture store, 20 hours a week, remote in the UK. Maximum 250 words. Include a short intro, responsibilities, required skills, and how to apply. No salary.
- Rewrite this for a customer so it is friendly and clear, keeping every fact: 'Pursuant to the terms outlined in your service agreement, the aforementioned maintenance window will necessitate a temporary suspension of platform availability commencing 02:00 UTC on Saturday and concluding no later than 06:00 UTC, during which time data integrity will be preserved but transactional operations will be unavailable.'
- Write a 5 step standard operating procedure for handling a credit card chargeback at a small ecommerce company. For each step give: what to do, who owns it (role, not name), the deadline relative to the chargeback notice, and the evidence to collect. Finish with two mistakes that lose disputes.
- A product costs us $26.00 and we sell it for $40.00. Our supplier is raising the cost by 12%. Calculate: the new cost, the price we need to keep the same 35% gross margin percentage, and the price we need to keep the same gross margin in dollars. Show the working step by step and round to cents.
Limitations
- Ten tasks, one run each, one day. A different task mix (long code, research synthesis, image analysis) could show a gap we did not see.
- We compared the models through Krater, not inside the ChatGPT app. ChatGPT adds its own tools, memory, routing and browsing on top of the models.
- No web browsing was enabled, which is why task 5 was partial for all three. That was a deliberate choice to test what the models know rather than what they can look up.
- Elapsed time is per comparison, not per model, because the three models ran concurrently.
- Prices, tiers and allowances are OpenAI's as of 17 September 2026 and will change.
Frequently Asked Questions
What is the difference between ChatGPT Plus and Pro?
Price and allowance. Plus is $20 a month with baseline usage. Pro is $100 a month for 5x the usage and 50 GPT-6 Pro messages a week, or $200 a month for 20x the usage and 200 GPT-6 Pro messages a week. Both Pro tiers have the same features; Plus does not include GPT-6 Pro in Chat.
Is ChatGPT Pro worth it?
It is worth it if you hit Plus limits regularly, run large Codex or deep research workloads, or need GPT-6 Pro for hard reasoning. On ten everyday business tasks, the Pro-only models scored the same as the Plus-level model while costing 4x to 13x more per answer, so for writing, spreadsheets and documents Plus is enough.
Can I still buy ChatGPT Pro for $200?
Not as of 10 September 2026. OpenAI paused new sign-ups and upgrades to the $200 tier. Existing $200 subscribers keep renewing, and the $100 tier is still available.
Does ChatGPT Plus include GPT-6 Pro?
No. GPT-6 Pro in Chat is available on Pro $100, Pro $200, Business and Enterprise. Plus includes GPT-6 Astra and GPT-5.6 in Chat, and GPT-6 Astra in ChatGPT Work and Codex.
Bottom Line
ChatGPT Pro is a usage upgrade, not a quality upgrade for everyday work. If you are rate limited on Plus or you live in Codex and deep research, Pro $100 is the tier you can buy today. If your week is emails, proposals, policies and spreadsheets, the model in Plus gave the same correct answers as the Pro-only models in our test, and you can prove that for your own tasks in a few minutes with a side-by-side run.