On 18 October 2026, OpenAI's Codex pricing page lists ChatGPT Plus at $20 a month and Pro "from $100 a month", with plans at $100, $200 or $500. The OpenAI API is billed per million tokens. The same page says Codex can also run on an API key, "charged at standard API rates." So there are really three ways to pay for coding: a plan, an API key, or both.
This post uses OpenAI's own pages fetched on 18 October 2026 (the Codex pricing page and the API pricing page). The openai.com pricing pages returned an access error when fetched, so nothing here comes from them.
What OpenAI publishes about plans
| Plan | Published price | Coding wording on the page |
|---|---|---|
| Free | $0 | Quick coding tasks |
| Go | $8 a month | Lightweight coding tasks |
| Plus | $20 a month | A few focused coding sessions per week |
| Pro | $100, $200 or $500 a month | "Pro plans currently have no five-hour limit." Ultrafast mode is on the $500 plan |
| Business | $20 a user on annual billing, $25 monthly | Shared workspace, admin controls, SAML SSO |
| Enterprise and Edu | Contact sales | Priority processing, SCIM, EKM, RBAC, Compliance API |
For Plus and standard Business seats, the page gives local message estimates per five-hour period. They depend on model: 5 to 45 for GPT-6 Astra, 15 to 160 for GPT-6.1 Sol, 15 to 150 for GPT-6 Sol, and 350 to 3,000 for GPT-6 Luna. OpenAI is explicit that "these estimates are not fixed message limits" and that the number of messages "depends on the model used, size and complexity of your tasks."
What the API charges
Per million tokens, Standard tier, short context (up to 272K input tokens), from the API pricing page:
| Model | Input | Cached input | Output |
|---|---|---|---|
| gpt-6-astra | $10.00 | $1.00 | $50.00 |
| gpt-6.1-sol | $2.00 | $0.10 | $10.00 |
| gpt-6-sol | $2.00 | $0.20 | $10.00 |
| gpt-5.3-codex | $1.75 | $0.175 | $14.00 |
| gpt-6-luna | $0.10 | $0.01 | $0.50 |
A scenario you can change
The plan page does not give tokens per message, so the comparison needs an assumption. These inputs are mine, not OpenAI data: one developer day with 3,000,000 cached input tokens, 350,000 fresh input tokens, 100,000 output tokens, over 22 working days.
prices = { # USD per MTok: fresh in, cached in, out
"gpt-6.1-sol": (2.00, 0.10, 10),
"gpt-6-sol": (2.00, 0.20, 10),
"gpt-5.3-codex": (1.75, 0.175, 14),
"gpt-6-astra": (10.00, 1.00, 50),
}
cached, fresh, out, days = 3.0, 0.35, 0.1, 22 # MTok per day
for m, (i, c, o) in prices.items():
day = fresh*i + cached*c + out*o
print(f"{m}: ${day:.2f} a day, ${day*days:.2f} a month")
The result is about $44.00 a month on gpt-6.1-sol, $50.60 on gpt-6-sol, $55.82 on gpt-5.3-codex and $253.00 on gpt-6-astra. Change the inputs and rerun. A table of this kind tells you which side of a plan price your own usage falls on. It does not tell you what any provider spends.
How to choose
- Plus is enough if you code in a few focused sessions a week, which is how OpenAI describes it.
- Pro is for daily heavy use. The page notes there is no five-hour limit on Pro, but it does not publish a token allowance.
- API key fits CI, shared environments and anyone who wants metered, per-token billing. The page calls it "great for automation in shared environments like CI."
- Business or Enterprise adds admin and security controls. Check data terms with your procurement team.
Safeguard works with ChatGPT through MCP, and MCP tool calls are unlimited, so tool usage does not become a second meter on top of your model bill.
Quick check
Can a Plus user keep coding after the plan allowance runs out?
The Codex pricing page says all users may run extra local chats using an API key, with usage charged at standard API rates.
