Codex Pricing 2026: Plus $20, Pro $100/$200/$500, API
How Codex is priced: Free $0, Go $8, Plus $20, Pro $100/$200/$500, credits, or pay-per-token API. Includes monthly estimates and Taiwan invoice notes.
Codex is bundled into ChatGPT plans: Free $0, Go $8, Plus $20, and Pro in three tiers: $100, $200, and $500. If you want to use an API key, you pay per token, calculated separately from the subscription.
Most people's choice is simple. If you write code only occasionally, Plus is enough. If you run it heavily every day, look at Pro. If you want to plug it into CI or your own agent, go with the API. The actual numbers follow.
This article was checked against OpenAI's Codex pricing page on September 30, 2026. Prices are in US dollars. Taxes and payment terms should be confirmed against your own account.
Plan Prices
| Plan | Monthly fee | Best for |
|---|---|---|
| Free | $0 | Trying a few small tasks |
| Go | $8 | Lightweight programming tasks |
| Plus | $20 | Focused coding sessions a few times a week |
| Pro | Starting at $100, in tiers of $100, $200, and $500 | Daily use with heavy usage |
| Business | $20 per person per month (billed annually), $25 if billed monthly, minimum 2 people | Team workspaces |
| Enterprise, Edu | Contact sales | Based on contract |
The three Pro tiers are different. According to OpenAI's earlier announcement, $100 gives 5 times the Plus usage, and $200 gives 20 times; only the $500 tier includes Ultrafast (explained below). The $200 tier has not accepted new subscriptions since September 10, 2026, but existing subscribers keep it.
There is one more change after DevDay: according to a report from The Next Web, starting October 30, Pro $200's usage in Codex will drop from 20 times Plus to 10 times Plus. Current subscribers keep their existing allowance until October 29, and there is also a one-time $2,500 usage credit. These are figures reported by the media; your account display is the final reference. We have also written about the impact: "Codex Pro $200 Usage Calculation Change Halves API Spend."
How Much the Subscription Covers
Plus and Business Standard are estimated by "how many messages can be sent every 5 hours." OpenAI gives a range, and it varies a lot depending on the model and task size:
| Model | Local messages per 5 hours (estimate) |
|---|---|
| GPT-6 Astra | 5 to 45 |
| GPT-6.1 Sol | 15 to 160 |
| GPT-6 Luna | 350 to 3,000 |
Pro has no fixed 5-hour cap. Changing one line of code versus having an agent read the entire project and then run tests can differ in consumption by dozens of times, so there is no fixed answer to "how much does one task cost."
After Your Plan Allowance Runs Out: Credits
Once you exceed your plan allowance, you can buy credits. The official rate card lists credits per million tokens as:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | 250 | 25 | 1,250 |
| GPT-6.1 Sol | 50 | 2.5 | 250 |
| GPT-6 Luna | 2.5 | 0.25 | 12.5 |
OpenAI also states that a typical GPT-5.6 Sol task uses roughly 5 to 30 credits. Fast mode uses 2 times the credits, and Ultrafast uses 6 times; Ultrafast is only available on Pro $500 and eligible Enterprise plans.
To convert: Astra's input rate of 250 credits corresponds to the API price of $10 per million tokens, so 1 credit is about $0.04. This is my derived estimate from comparing the table above with API prices; the official page does not state it directly, so confirm against the credit price in your account before purchasing.
GPT-5.5 will be removed from ChatGPT and Codex on all plans on October 14, 2026. If you are still using it, switch models first.
Using an API Key: The Most Direct Calculation
API costs are separate from the subscription. Per million tokens:
| Model | Input | Output | Cache read |
|---|---|---|---|
| GPT-6 Astra | $10 | $50 | Per official pricing |
| GPT-6.1 Sol | $2 | $10 | $0.10 |
| GPT-6 Luna | $0.1 | $0.5 | Per official pricing |
These are the prices listed in the BazaarLink model catalog. The Ultrafast API price is 6 times the standard price; for example, Astra Ultrafast is $60 input and $300 output.
Here is an example calculation. Assume 50 model requests per day, each with 2,000 input tokens and 500 output tokens, for 30 consecutive days. That totals 3 million input and 750 thousand output tokens, without caching:
| Model | Calculation | One month |
|---|---|---|
| GPT-6.1 Sol | 3 × $2 + 0.75 × $10 | $13.5 |
| GPT-6 Astra | 3 × $10 + 0.75 × $50 | $67.5 |
| GPT-6 Luna | 3 × $0.1 + 0.75 × $0.5 | About $0.68 |
These are 50 model requests, not 50 agent tasks. An agent task often calls the model many times, rereads context, and retries, so for real figures, export your request logs and calculate from those.
How to Choose
| Your situation | Recommendation |
|---|---|
| Occasional use, a few times a week | Plus, check whether it's enough first |
| Fixed daily use, allowance often runs out | Pro $100; look at higher tiers if you need more |
| Only need one model and connecting it to your own code | API; run the calculation above once first |
| CI, batch jobs, self-built agents | API; OpenAI lists automation as a use case for API keys |
| Irregular usage (seasonal peaks and lulls in freelance work) | Subscription as a base, use credits or API for overage, cancel the subscription in slow seasons |
If you only use the official Codex tools (CLI, IDE, web, desktop) and sign in with ChatGPT, choose an official plan; if you only need model API access, compare API prices. Official cloud workflows cannot be fully replaced by switching to a compatible API, so check the tool documentation before connecting.
Expense Reports and Invoices
ChatGPT subscriptions and the API use different billing entry points. OpenAI has an explanation of Taiwan VAT and eGUI, but an eGUI is only issued when VAT was charged on that particular transaction; entering a unified business number does not guarantee one. Whether overseas charges can be recognized as expenses and whether input tax can be deducted are separate questions, so leave them to your accountant. For the full process, see "AI API Expense Report Process" and "ChatGPT Unified Business Number and Invoices."
Jilian Technology Co., Ltd. can issue two-part or three-part electronic uniform invoices to buyers who qualify under Taiwan transaction rules. When buying API services from BazaarLink, confirm the title, unified business number, and invoicing details before paying; this invoice corresponds to our API service and is not issued on behalf of a ChatGPT subscription. BazaarLink does not resell official Codex subscriptions and does not guarantee that official Codex login methods and features can be directly replaced.
If you want to cancel a subscription in a slow season, the cancellation entry point depends on your purchase channel; deleting the app does not cancel the subscription. See "Cancellation and Refund Guide."
FAQ
Do I have to subscribe to ChatGPT Plus to use Codex?
No. Codex is bundled into ChatGPT plans: Free $0 and Go $8 also work, though each comes with limited usage. Plus is $20, and Pro has three tiers: $100, $200, and $500. Pricing follows OpenAI's official Codex pricing page, checked on September 30, 2026.
How much does a single Codex task cost?
There is no fixed unit price. OpenAI says a typical GPT-5.6 Sol task uses roughly 5 to 30 credits. Based on Astra's credit rates compared with its API prices, 1 credit is about $0.04 (this is a derived estimate), so roughly $0.2 to $1.2. Task size, context, and retries all move the number up or down.
Does the Plus monthly fee include API costs?
No. Subscription usage, credits, and per-token API billing are three separate calculations, so API costs cannot be converted directly into the Plus monthly fee.
What is Ultrafast, and how much does it cost?
It is a faster speed tier, up to about 8 times faster in Codex. It uses 6 times the standard credits, and its API price is also 6 times the standard price (for example, Astra Ultrafast is $60 input and $300 output). Currently it is only available on Pro $500 and eligible Enterprise plans.
Roughly how much does a month of API usage with GPT-6.1 Sol cost?
Assuming 50 model requests per day, 2,000 input and 500 output tokens per request, for 30 consecutive days, and not counting caching, the estimate is about $13.5 (3 × $2 + 0.75 × $10). This is 50 model requests, not 50 agent tasks, so calculate from your actual request logs.
Does using the BazaarLink API mean I am buying an official Codex subscription?
No. BazaarLink provides model API services and does not resell official Codex subscriptions. Login methods, cloud workflows, and tool features should be checked in each tool's documentation; it is not guaranteed that every feature of official Codex can be replaced.
If my freelance Codex costs have a Taiwan uniform invoice, can I always deduct them?
Not by looking at the document title alone. Keep the real transactions, purpose, payment, and project records, and confirm expense recognition and input tax deduction separately with your accountant. When you buy API services from Jilian Technology that qualify under Taiwan transaction rules, we can issue a two-part or three-part electronic uniform invoice based on the buyer's details.
TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API