Which plan should you pick?
There’s no Codex subscription. Codex comes with ChatGPT plans, and you sign in with your ChatGPT account. The alternative is an OpenAI API key, billed per token.
- Free and Go ($8/month) get Codex in the ChatGPT desktop app with GPT-6 Luna at standard speed, as it rolls out. Neither includes Codex Cloud. They’re fine for trying Codex, not for daily coding.
- Plus ($20/month) is the first plan with the full product: Codex on the web, in the CLI, the IDE extension and iOS. You get GPT-6.1 Sol and GPT-6 Luna, automatic code review and Slack. You can buy credits when you hit the limits. For most developers it’s the starting point.
- Pro ($100, $200 or $500/month) comes in three tiers with increasing usage. Pro currently has no 5-hour limit, though weekly limits may apply. Only Pro 500 includes Ultrafast mode. Pro 200 is open to new subscribers again, but new subscriptions get a lower allowance than existing ones, which keep theirs through October 29, 2026.
- An API key suits scripts, CI and the SDK. You pay per token and don’t get cloud features. See paying per token below.
Go, Plus and Pro are billed monthly only.
How the usage limits work
Codex limits work in 5-hour windows, and weekly limits can apply as well. The allowance is shared with ChatGPT Work, Excel, Word, PowerPoint and Workspace Agents, but not with regular ChatGPT chat. Cloud tasks use more of it than local messages do.
For Plus and Business Standard seats, OpenAI publishes estimated local messages per 5 hours:
| Model | Local messages per 5 hours |
|---|---|
| GPT-6 Astra | 5–45 |
| GPT-6.1 Sol | 15–160 |
| GPT-6 Sol | 15–150 |
| GPT-6 Luna | 350–3,000 |
These are estimates, not fixed caps; bigger tasks use more. OpenAI doesn’t publish numbers for Free, Go or the Pro tiers, or for code reviews. Enterprise and Edu seats without flexible pricing get Plus-level limits.
Speed costs extra. Fast mode uses 2.5 times the included usage (or twice the credits). Ultrafast, on GPT-6 Astra and GPT-6.1 Sol only, uses 8 times the included usage and then 6 times the credits.
When you reach a limit, Codex lets the current turn finish. After that you can:
- wait for the window to reset;
- buy credits (Plus and Pro);
- switch to an API key;
- upgrade.
Credits: paying for more Codex
Credits extend Codex past your plan’s limits. They last 12 months, aren’t refundable and can auto-reload. Only a limited group of Free and Go users can buy them.
OpenAI doesn’t publish one credit price. Its student offer values 2,500 credits at $100, or 4 cents each, but it says purchase prices vary by plan. Credits are charged per million tokens:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | 250 | 25 | 1,250 |
| GPT-6.1 Sol | 50 | 2.5 | 250 |
| GPT-6 Sol | 50 | 5 | 250 |
| GPT-6 Luna | 2.5 | 0.25 | 12.5 |
Credits per million tokens. OpenAI says a typical task on GPT-5.6 Sol uses 2–15 credits. At 4 cents a credit, that’s roughly $0.08 to $0.60 a task. Unlike the API, credits have no cache-write charge, and auto-review safety checks are free.
Paying per token with an API key
Sign in with an OpenAI API key instead of ChatGPT and Codex bills your API account at standard rates. This works in the CLI, the SDK and the IDE extension, but not for cloud features. API-key sessions also still have access to models that were retired from ChatGPT sign-in, such as GPT-5.5.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | $10 | $1 | $50 |
| GPT-6.1 Sol | $2 | $0.10 | $10 |
| GPT-6 Sol | $2 | $0.20 | $10 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
| GPT-5.5 (under 272K context) | $5 | $0.50 | $30 |
US dollars per million tokens, standard tier. Batch is half price and Fast is double. Ultrafast is $60 input and $300 output on Astra, and $12 and $60 on GPT-6.1 Sol. The API also charges for cache writes.
The CLI can also run local models with --oss, through Ollama or LM Studio, at no cost from OpenAI.
Business, Enterprise and Edu
- Business needs at least 2 seats. A Standard seat costs $25 per user per month, or $20 billed annually. A Premium seat costs $125, or $100 annually, and has higher limits. Workspaces buy pooled credits in billing, with auto-reload and monthly limits per seat type or per user. Admins can set the default model, reasoning level and speed, and control Fast mode and cloud access.
- Credits-only Codex seats had no seat fee. New Business workspaces haven’t been able to add them since June 24, 2026; existing ones keep working.
- Enterprise and Edu are priced by sales. Seats get Plus-level limits, or no fixed limits with flexible pricing, which uses a contract-level shared credit pool with overage limits and alerts. Cloud access is off by default for Enterprise.
Discounts
- Annual billing: Business only ($20 instead of $25 per Standard seat).
- Students in the US and Canada can get $100 in Codex credits (2,500 credits).
- Open-source maintainers can apply for six months of Pro plus API credits.
- OpenAI hasn’t published a startup discount for Codex.
Claude Code
Cursor
GitHub Copilot
Gemini CLI
OpenCode