You want to use Codex, and the pricing page answers the wrong question. It tells you the plan prices, then hides what each plan actually buys you behind ranges like “15 to 160 messages” and a credit table with no dollar figure on it. So you end up guessing: is $20 enough, do you need $100, and why would anyone pay $500?
I pay for Codex next to Claude Code and push a lot of work through both, so I have had to answer this with my own money. Below is every Codex price as of 11 November 2026, checked against OpenAI’s own page, what the plans really mean in use, the maths that makes the $200 tier the best published value per dollar, and a one-week test that tells you which plan you need before you upgrade.
On this page
The short answer
Codex has no price of its own. It is included in every ChatGPT plan, from Free at $0 to Pro at $500 a month, and the plan you pick only decides how much Codex you get and how fast. You can also skip the plans entirely and pay per token with an OpenAI API key.
My ranked picks:
- Pro $200 if Codex is a daily tool and you run agents for hours. It gives 20 times Plus usage for 10 times the price, the best value per dollar of any tier where OpenAI publishes the multiple.
- Plus $20 if you code with Codex a few hours a week, or if it is your second tool behind something else. Most people should start here.
- Pro $100 only if Plus stops you most working days and $200 is genuinely out of budget. It is the same usage per dollar as Plus, just more of it.
Go and Free are for trying Codex, not for working in it. The $500 tier makes sense only if you need Ultrafast mode.
Codex pricing at a glance (November 2026)
Every price and feature below comes from OpenAI’s Codex pricing page, checked on 11 November 2026. The Pro usage multiples come from OpenAI’s help center article on ChatGPT Pro.
| Plan | Price | What you get in Codex |
|---|---|---|
| Free | $0 | Limited Codex on quick tasks, GPT-6 Luna at standard speed in the desktop app |
| Go | $8/month | Same light access as Free, for small tasks |
| Plus | $20/month | Codex on web, CLI, IDE extension and iOS, cloud code review and Slack, GPT-6.1 Sol and GPT-6 Luna, can buy extra credits |
| Pro | $100, $200 or $500/month | $100 is 5x Plus usage, $200 is 20x Plus usage, $500 adds Ultrafast mode; no five-hour limit right now |
| Business | $20/user/month yearly (2+ users), $25 monthly | Codex plus admin controls, can buy workspace credits |
| Enterprise and Edu | Contact sales | Credit-based usage, no fixed rate limits on flexible pricing |
| API key | Pay per token | Codex CLI, SDK and IDE on API rates, no cloud features such as GitHub code review |
Two details on that page change the maths more than the prices do. First, ChatGPT Work and Codex share one usage pool, so the deep research and agent work you do in ChatGPT eats the same allowance your coding does. Second, weekly limits can apply on top of the five-hour windows, and OpenAI does not publish the weekly number.
If an older guide shows a different ladder, it is out of date. OpenAI added the $100 Pro tier on 9 April 2026, between Plus and the $200 plan, and the pricing page now also lists a $500 tier with Ultrafast mode. Guides that still describe Codex as Free, Plus and a single $200 Pro plan, or price it by the old codex-1 model, predate both changes.
What $20 a month actually buys you
The most common complaint about Codex pricing is that nobody tells you what Plus really gets you. OpenAI does publish it, as ranges of local messages per five-hour window on Plus:
| Model | Messages per 5 hours on Plus |
|---|---|
| GPT-6 Astra (the most expensive) | 5 to 45 |
| GPT-6.1 Sol | 15 to 160 |
| GPT-6.1 Terra | 35 to 285 |
| GPT-6 Luna (the cheapest) | 350 to 3,000 |
The range is wide because a “message” is not a fixed unit. A one-line fix and an agent that reads half your repository both count as one message, but the second burns far more of your allowance. Cloud tasks can use more than local ones.
On Plus, the model you pick matters more than the plan you pay for. The same five hours gets you as few as 5 Astra messages or as many as 3,000 Luna messages. If you leave Codex on its biggest model for everything, Plus will feel tiny. If you drop to Luna or Terra for routine work, it stretches a long way.
To see where you stand, run /status in an active Codex CLI session, or open the Codex usage dashboard, which shows your remaining five-hour and weekly allowance and when each resets.
The Pro tiers: why $200 is the best value per dollar
Pro comes in three prices. The useful way to compare them is usage per dollar against Plus:
| Plan | Price vs Plus | Usage vs Plus | Usage per dollar vs Plus |
|---|---|---|---|
| Plus $20 | 1x | 1x | 1x |
| Pro $100 | 5x | 5x | 1x |
| Pro $200 | 10x | 20x | 2x |
| Pro $500 | 25x | Not published | Buy it for Ultrafast, not for value |
Pro $100 is not a discount. It is five Plus plans in one bill. Pro $200 is the only tier that doubles what each dollar buys.
That is the whole case for the $200 tier. If you regularly hit the Pro $100 ceiling, Pro $200 doubles what each dollar buys. Credits are the exception: if you only run out once or twice a month, topping up is cheaper than any upgrade, because you pay only for the overflow. The $100 tier is still the right call if you are sure you will never need more than five times Plus, because then you are not paying for headroom you will not use.
All Pro tiers currently drop the five-hour window. That matters more than the multiple for some people: on Plus, a long agent run can stall halfway through a working afternoon, while on Pro you only run into the weekly limit.
Credits and Fast mode
When you run out, Plus and Pro users can buy ChatGPT credits instead of upgrading. OpenAI publishes how many credits each model uses per million tokens, but not a dollar price per credit on the pricing page; the price depends on your plan and shows up when you buy. The ratios are still worth knowing, because they tell you how to stretch whatever you have:
| Model | Credits per 1M input tokens | Credits per 1M output tokens | Cost vs GPT-6.1 Sol |
|---|---|---|---|
| GPT-6 Astra | 250 | 1,250 | 5x |
| GPT-6.1 Sol | 50 | 250 | 1x |
| GPT-6.1 Terra | 25 | 125 | 0.5x |
| GPT-6 Luna | 2.5 | 12.5 | 0.05x |
GPT-6 Luna costs one twentieth of GPT-6.1 Sol per token, so moving routine work to Luna makes the same credits last about 20 times longer. Output tokens cost five times input on every GPT-6 model in that table, which is why agents that write a lot of code drain credits faster than ones that mostly read.
Speed costs extra too. Fast mode uses 2.5 times your included allowance, and Ultrafast uses 8 times. Turn them on when you are waiting on something, not by default.
Codex API pricing: what pay-per-token really costs
The API key route skips the plans: you sign Codex in with an API key and pay standard OpenAI API rates per token. You lose the cloud features, including GitHub code review and Slack. These are the standard rates from OpenAI’s API pricing page, checked on 11 November 2026, for prompts up to 272K tokens:
| Model | Input per 1M tokens | Cached input per 1M | Output per 1M tokens |
|---|---|---|---|
| GPT-6 Astra | $10.00 | $1.00 | $50.00 |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 |
| GPT-6.1 Terra | $1.00 | $0.05 | $5.00 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
Here is a worked example so you can run your own numbers. It is an illustration, not my bill. Say one agent session on GPT-6.1 Sol reads 2 million input tokens and writes 200,000 output tokens:
Cost of one session on the API (GPT-6.1 Sol, no cache hits)
input: 2.0M tokens x $2.00 = $4.00
output: 0.2M tokens x $10.00 = $2.00
total per session = $6.00
10 sessions a day x 20 working days = 200 sessions
200 x $6.00 = $1,200 a month on the API
versus $200 a month on Pro $200
Cached input is 20 times cheaper than fresh input on Sol, and agents re-read the same files constantly, so real bills often land well below the no-cache figure. Even so, a full-time agentic workload on the API usually costs several times a flat plan. OpenAI pitches the API key for automation in shared environments like CI, and that is where I would use it. For a person coding all day, it is almost always the expensive option, for the reason in the next section.
What I actually pay, and why the flat plan wins for heavy users
My AI bill is roughly $500 to $600 a month across providers. Claude is about $400 to $500 of that, a personal Max 20x seat plus two team seats I keep for the connectors. Codex plus a couple of smaller model bills come to about $100. I have also paid for OpenAI’s top plan while testing it properly, so the verdicts here are not from a free trial.
The number that matters is the usage behind that bill. I push about 15 to 20 billion tokens a month through these tools. At API list prices that would cost $15,000 to $20,000 a month; on flat plans I pay $500 to $600. OpenAI and Anthropic both sell a $200 tier for heavy users, and that $200 is a subsidy only a lab that owns the model can afford to hand a heavy user. My read, and it is an opinion: even after stripping out their margins, the labs lose at least $5,000 a month on a user like me, and the subscription model is not sustainable at this price.
That is also why I cancelled Cursor. Cursor does not own a flagship model, so it resells other labs’ tokens. On its $100 to $200 plans I got roughly $1,000 of API credit, which against what $200 buys on a first-party plan felt like about 15 times less usable work. The lesson carries straight over to Codex: if you are a heavy user, buy usage from the company that owns the model, on a flat plan, not per token.
The same logic made me drop a cheaper option. I paid for the GLM coding plan and liked it, but saving $100 to $200 a month was not worth working on a model I knew was not the best. For light, long jobs I keep the DeepSeek API instead, so that grunt work does not burn my premium limits.
Which Codex plan should you buy?
| If you are… | Buy | Why |
|---|---|---|
| Trying Codex for the first time | Free or Go | Enough to see if the workflow suits you, not enough to work in it |
| Coding with Codex a few hours a week | Plus $20 | Usually enough if you use smaller models for routine work |
| Using Codex as your second tool behind Claude Code or Cursor | Plus $20 | Overflow and backend work fit inside Plus for most people |
| Running agents for hours most days | Pro $200 | 2x usage per dollar and no five-hour window |
| Stopped by Plus most days but capped at about $100 | Pro $100 | Five times Plus, no five-hour window, same value per dollar |
| Needing the fastest possible output | Pro $500 | The only tier with Ultrafast mode |
| A team of 2 or more | Business | Admin controls, $20 per user on yearly billing |
| Running Codex in CI or scripts | API key | Billed per token, built for shared automated environments |
Before you upgrade, run this one-week test. It costs nothing and it stops you buying headroom you do not need.
Codex upgrade test (one working week)
Every day, at the end of the day:
1. Run /status in the Codex CLI, or open chatgpt.com/codex/settings/usage
2. Note: did the five-hour limit stop you today? (yes/no)
3. Note: weekly allowance left (%)
4. Note: which model you used most
At the end of the week:
- Five-hour limit stopped you on 0 to 1 days -> stay on Plus
- Stopped you on 2 or more days -> first switch routine
work to Luna or Terra
for one more week
- Still stopped on 2 or more days after that -> Pro $100
- Weekly allowance gone before day 5 on Pro -> Pro $200
- You only run out once or twice a month -> buy credits, do not upgrade
The middle step is the one people skip. Changing the default model is free, and on the published rates it can be worth far more than the jump from Plus to Pro.
How I split work so neither subscription is wasted
I use Claude about 70% of the time and Codex for the rest, both from their terminal CLIs, with around 20 sessions open at once. Paying for two plans only makes sense if each one gets the work it is best at, so the split is deliberate:
- Claude Code gets anything with UI or UX, and whole features end to end. When I brief a full feature by voice, it goes to Claude, because it handles frontend and backend together and is far more intuitive about interfaces. Codex is weak at interface work.
- Codex gets complex backend work. In my experience OpenAI’s models have been better at hard backend logic since GPT-4o.
- Codex also gets the overflow when my Claude limits run out mid-task, which on its own justifies a Plus plan for anyone who codes all day.
- A cheap API model gets long, trivial jobs, so neither premium allowance is spent on them.
If you run both tools, write the split down once so you stop deciding it every time. Here is the version I would put in a shared notes file or at the top of an AGENTS.md (the AGENTS.md format Codex reads):
Routing rules for this repo
- UI, UX, design and end-to-end features: Claude Code
- Complex backend logic, data and algorithm work: Codex
- When one tool hits its usage limit mid-task: hand the task to the other
- Long, simple, high-volume jobs: cheapest model available, never the premium one
- Default Codex model for routine edits: a small model (Luna or Terra), switch up only when it fails
The last line is the one that saves money. It turns the credit table above into a habit.
Two things make this split easier for me. When I hand a whole feature to AI, I brief it by voice through Contextli, the voice-to-AI app I built so the context on my screen goes into the prompt without copying and pasting. And on the Claude Code side, my free Claude skills are plain instruction files you can install in one command.
What Reddit gets right and wrong about Codex pricing
“Codex is worth $1,000 a month, so Plus users should stop complaining.” Mostly right about the value. At API prices, heavy usage really is worth many times the subscription, and my own numbers are an extreme version of that. It is wrong as advice, though: if Plus stops you every afternoon, the fix is the model you default to or the tier, not gratitude.
“I rarely hit the five-hour limit, I hit the weekly one. Will Pro help?” Yes, but only through the multiple. Pro removes the five-hour window, and OpenAI does not say it removes weekly limits. If the weekly cap is your wall, Pro $100 gives you five times as much and Pro $200 twenty times.
“OpenAI does not tell you what $20 actually gets you.” It does, as message ranges per model, and the table above has them. The ranges are wide because message size varies, which is why the one-week test beats reading the table.
“I only upgraded for Codex; otherwise Plus is plenty.” That is the honest pattern. For chat, research and writing, Plus is enough for most people. The upgrade case is almost always agentic coding, and remember that ChatGPT Work shares the same pool, so heavy agent use in ChatGPT counts against Codex too.
FAQ
How much does Codex cost?
Codex is included in ChatGPT plans: Free ($0), Go ($8/month), Plus ($20/month), Pro ($100, $200 or $500/month) and Business ($20 per user per month on yearly billing, $25 monthly). Enterprise and Edu are priced by OpenAI’s sales team. You can also use Codex with an API key and pay standard API rates per token.
What is the price of ChatGPT Pro?
ChatGPT Pro comes in three tiers: $100, $200 and $500 a month. The $100 tier gives five times Plus usage, the $200 tier gives twenty times Plus usage, and the $500 tier adds Ultrafast mode. All three currently have no five-hour limit in Codex.
Does Codex require a subscription?
No. You can use it on the Free plan with limited access, or sign it in with an OpenAI API key and pay per token. For regular daily coding, a paid ChatGPT plan is usually much cheaper than the API key route.
Is Codex free with ChatGPT Go?
Go includes limited Codex access with GPT-6 Luna at standard speed in the desktop app. It is enough for light tasks. The full set of Codex surfaces, the CLI, IDE extension, web and cloud code review, starts at Plus.
Is Codex cheaper than Claude Code?
At the same price points they are close. Here is how the individual plans line up, from OpenAI’s pricing page and Claude’s pricing page, both checked on 11 November 2026:
| Price | Codex (ChatGPT) | Claude Code (Claude) |
|---|---|---|
| $0 | Free: limited Codex | Free: no Claude Code |
| $20 | Plus: 5-hour and weekly limits | Pro: 5-hour and weekly limits ($17 a month yearly) |
| $100 | Pro: 5x Plus, no 5-hour limit | Max: 5x Pro, 5-hour and weekly limits |
| $200 | Pro: 20x Plus, no 5-hour limit | Ultra: 10x Pro weekly, no 5-hour limit |
| $500 | Pro: adds Ultrafast mode | No individual plan |
The difference is in how far each plan stretches for your kind of work. My breakdown of Claude Code pricing covers the Anthropic side, and my Codex vs Claude Code verdict covers which one I would keep.
What happens when I hit my Codex limit?
You wait for the window to reset, buy ChatGPT credits (Plus and Pro), upgrade, or switch to a smaller model, which uses far less of your allowance. Business and Enterprise workspaces on flexible pricing can buy workspace credits.
Where this comes from
I am Junaid Khalid, the founder of Ertiqah, and I build and run my own portfolio of software products, with all of the coding and most of the back office done through AI tools in a terminal. I push about 15 to 20 billion tokens a month through them. In the last year I have paid for Codex, Claude Code, Cursor, the GLM coding plan and the DeepSeek API, kept Codex, Claude Code and DeepSeek, and cancelled the others for the reasons above. The plan verdicts in this article are what I would buy with my own money, not a reading of a feature list. Prices and limits were last verified against OpenAI’s and Anthropic’s own pricing pages on 11 November 2026, and they change often, so check them before you buy.