TL;DR (October 10, 2026): Codex comes with ChatGPT plans: Free $0 and Go $8/month (GPT-6 Luna in the desktop app, subject to rollout), Plus $20/month (Codex on web, CLI, IDE and iOS with GPT-6.1 Sol and GPT-6 Luna), Pro from $100/month (plans at $100, $200 or $500, with no five-hour limit for now), and Business $20 per user per month billed annually ($25 monthly). On Plus, OpenAI estimates 15 to 160 local GPT-6.1 Sol messages per five hours, or 350 to 3,000 on GPT-6 Luna. Past your limit you can buy credits, or run Codex with an API key at API token prices. GPT-5.5 leaves Codex on October 14, 2026.
Codex pricing is two systems glued together. Most people use it through a ChatGPT subscription, where usage is a rolling allowance measured in messages and reset windows. Anyone can also point Codex at an OpenAI API key, where every token is billed at API rates. Since ChatGPT Work and Codex now share one allowance, the plan you pick for chat also decides how much coding agent time you get. This page lays out each plan, the usage estimates OpenAI publishes, the credit rates that apply after you run out, and the October model changes.
Codex plans and prices as of October 2026
| Plan | Price | Codex access | Models |
|---|---|---|---|
| Free | $0 | Desktop app, quick coding tasks | GPT-6 Luna at Standard speed, subject to rollout |
| Go | $8 / month | Desktop app, lightweight tasks | GPT-6 Luna at Standard speed, subject to rollout |
| Plus | $20 / month | Web, CLI, IDE extension, iOS, cloud code review and Slack | GPT-6.1 Sol and GPT-6 Luna |
| Pro | $100, $200 or $500 / month | Everything in Plus; Ultrafast mode on Pro $500 | As Plus, plus Ultrafast on $500 |
| Business | $20 / user / month annual, $25 monthly (2+ users) | Desktop and mobile, larger cloud VMs, admin controls, SAML SSO | As Plus |
| Enterprise and Edu | Contact sales | Priority processing, SCIM, RBAC, audit logs via the Compliance API | GPT-6.1 Sol off by default until an admin enables it |
| API key | API token prices | CLI, SDK and IDE extension only; no cloud features | Whatever API models your key can use |
From OpenAI's Codex pricing page, read October 10, 2026. Prices in USD.
How many Codex messages you get per five hours
OpenAI does not publish a fixed message cap. It publishes ranges for local messages per five-hour window on Plus and Standard Business, and is clear that they are estimates rather than limits:
| Model | Plus (local messages per 5 hours) | Standard Business (local messages per 5 hours) |
|---|---|---|
| GPT-6 Astra | 5 to 45 | 5 to 45 |
| GPT-6.1 Sol | 15 to 160 | 15 to 160 |
| GPT-6 Sol | 15 to 150 | 15 to 150 |
| GPT-6 Luna | 350 to 3,000 | 350 to 3,000 |
The spread inside each range is wide because the cost of a message depends on context size, reasoning effort, tool calls, retrieval and caching. A short question about one function sits at the top of the range; a long session with a big repository loaded sits at the bottom. A few more rules from the same page:
- Pro has no five-hour limit at the moment. Weekly limits may still apply on any plan.
- Local messages and cloud chats share one allowance, and cloud tasks may use more of it than local messages.
- ChatGPT Work uses the same allowance as Codex, and so does ChatGPT for Excel on Plus and Pro.
- Image generation uses included limits about 3 to 5 times faster on average, and is not available on Free.
- Code review through GitHub (tagging
@Codexon a pull request, or automatic reviews) counts as Code Review usage. Reviews run locally count toward general limits. - Enterprise and Edu without flexible pricing get the same per-seat limits as Plus for most features; with flexible pricing there are no fixed rate limits and usage scales with credits.
You can check what is left in the Codex usage dashboard, or run /status inside a Codex CLI session.
Fast and Ultrafast burn the allowance faster
Speed costs usage. Relative to Standard speed on the same model:
| Speed mode | Included subscription usage | Purchased credits and Enterprise pay-as-you-go |
|---|---|---|
| Fast | 2.5x | 2x |
| GPT-6 Astra Ultrafast | 8x | 6x |
| GPT-6.1 Sol Ultrafast | 8x | 6x |
Fast mode works on GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol and GPT-6 Luna, and you toggle it with /fast in the CLI. Ultrafast is only on Pro $500 and eligible Enterprise and Edu plans; OpenAI says other self-serve plans cannot use it "even with purchased credits". On Pro $500 it draws on included usage first, then credits. Note the asymmetry: Fast costs 2.5x of your included allowance but only 2x of purchased credits, so once you are paying for credits anyway, the penalty for speed is a little smaller.
Credit rates after you hit the limit
Plus and Pro users who hit their limit can buy ChatGPT credits instead of upgrading. Business, Edu and Enterprise plans with flexible pricing buy workspace credits. Credits are charged per million tokens at these Standard-speed rates:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | 250 credits | 25 credits | 1,250 credits |
| GPT-6.1 Sol | 50 credits | 2.5 credits | 250 credits |
| GPT-6 Sol | 50 credits | 5 credits | 250 credits |
| GPT-6 Luna | 2.5 credits | 0.25 credits | 12.5 credits |
| GPT-5.6 Sol | 100 credits | 10 credits | 500 credits |
| GPT-5.6 Terra | 50 credits | 5 credits | 300 credits |
| GPT-5.6 Luna | 5 credits | 0.5 credits | 30 credits |
| GPT-5.5 | 125 credits | 12.50 credits | 750 credits |
Codex credit billing has no separate cache-write charge. OpenAI's page says a typical GPT-5.6 Sol task may use 2 to 15 credits, and that credit purchase prices and discounts depend on your plan or agreement.
One thing you can check yourself: if you put this table next to the API price list, every GPT-6 rate is exactly 25 credits per US dollar of API list price. GPT-6.1 Sol is $2 input and $10 output on the API, and 50 and 250 credits here. That makes it easy to compare the two routes once you know what a credit costs on your plan. OpenAI's own warning still applies: API token prices are not a way to estimate how many tasks your included subscription covers.
Codex with an API key
Signing in with an API key turns Codex into a pay-per-token tool. Usage follows API pricing, the ChatGPT credit multipliers for Fast and Ultrafast do not apply, and cloud features such as GitHub code review and Slack are not available. It suits CI and other shared automation, where a personal subscription is the wrong account to bill. Any user can also run extra local chats with an API key after hitting their plan limit. The per-token prices are in our GPT-6 API pricing page.
The trade-off is the one every AI coding tool now has: a subscription caps your bill but rations your work, while API billing never stops you but never caps the cost. We compared how Claude Code, Cursor, Copilot and others handle this in AI coding tool pricing in 2026, and the cost of an agent that keeps going in the $1,400 hour.
October 2026 model changes
- GPT-5.5 retires on October 14, 2026 from ChatGPT, ChatGPT Work and Codex on all plans. The API is not affected. OpenAI tells Plus, Pro, Business, Enterprise and Edu users to switch to GPT-6 Sol, and Free and Go users to GPT-6 Luna. Saved model settings and workspace defaults that name
gpt-5.5need updating. - GPT-6.1 Sol is rolling out on Plus, Pro, Business, Enterprise and Edu, and OpenAI recommends it for complex coding and agent work. In ChatGPT it appears in Work and Codex, not in Chat.
- GPT-5.6 Sol, Terra and Luna remain available during the rollout, at unchanged credit rates.
Making the allowance last
OpenAI's own advice is concrete: keep prompts tight, give the agent only the files it needs, keep AGENTS.md small (or nest several smaller ones), and disable MCP servers you are not using, since each one adds context to every message. The model choice matters most. Moving routine work from GPT-6.1 Sol to GPT-6 Luna changes the Plus estimate from 15 to 160 messages to 350 to 3,000 per window.
For teams, the harder problem is knowing who used what. Workspace credits are pooled, so the bill shows a total. If you need per-developer or per-project cost, you have to record it where the work starts. UsageBox meters the usage events your own tooling emits; it does not read OpenAI's usage dashboard.
Sources
Read on October 10, 2026: the OpenAI Codex pricing page, the ChatGPT Work and Codex documentation pages on Speed and on Models (including the GPT-5.5 retirement notice), and OpenAI's API pricing page for the per-token comparison.