Which ChatGPT plans include Codex? All of them — here's the breakdown

Why this is suddenly confusing

Three changes landed within three months. In July 2026 Codex was folded into ChatGPT (it joined the ChatGPT desktop app on July 9) and became part of every ChatGPT plan, metered by the plan’s allowance. In September the GPT-6 generation replaced the GPT-5.6 tiers: Astra, the top model, on September 3, and Sol and Luna on September 22. There is no GPT-6 Terra. Then at DevDay on September 29, GPT-6.1 Sol replaced GPT-6 Sol as the model OpenAI recommends for complex work, offering “near-Astra performance for complex work at a lower cost than Astra,” and Pro became three plans at $100, $200 and $500 with no five-hour limit.

OpenAI’s models page still lists GPT-6 Sol and keeps the GPT-5.6 tiers selectable during the rollout, and GPT-5.5 retired from ChatGPT, Work and Codex on every plan on October 14, 2026 (the API keeps it). Answers written before DevDay describe a Pro plan that no longer exists in that form.

The plans, as of October 2, 2026

• Free ($0): quick coding tasks; GPT-6 Luna at standard speed in the desktop app, subject to rollout.
• Go ($8/mo): lightweight coding tasks; the same Luna access as Free.
• Plus ($20/mo): roughly 5–45 Astra, 15–160 GPT-6.1 Sol, 15–150 GPT-6 Sol or 350–3,000 Luna messages per five-hour window; credits can extend it.
• Pro ($100, $200 or $500/mo, billed monthly only): no five-hour limit, more included usage at each step, and Ultrafast on the $500 plan only, for GPT-6 Astra and, since October 8, GPT-6.1 Sol. OpenAI publishes no message estimates for Pro. The $200 plan is open to new subscribers again with a lower allowance than before; OpenAI said anyone with an active $200 subscription at any point from September 22 to 10 a.m. Pacific on September 29 would keep the old allowance through October 29, 2026, even if they cancelled and subscribed again. Pro also includes a dot, OpenAI’s always-on agent, outside the EEA, the UK and Switzerland at launch.
• Business, Standard seat ($20/user/mo billed annually, $25 monthly): Plus-level estimates, extendable with workspace credits. A workspace needs at least two seats.
• Business, Premium seat ($100/user/mo billed annually, $125 monthly): five times a Standard seat’s usage, no five-hour limit, weekly resets. When OpenAI says “Business Premium,” as it did launching dots, it means this seat: a seat type inside a Business workspace, not a separate plan. Seats can be mixed, so by arithmetic the cheapest workspace with one Premium user is one Premium and one Standard seat, $120 a month on annual billing.
• Enterprise and Edu (through sales): with flexible pricing, no fixed rate limits, as usage scales with credits; without it, the same per-seat limits as Plus for most features.

The figures are from OpenAI’s Codex pricing page and its help articles on Pro tiers and Business seats. In ChatGPT, GPT-6.1 Sol, GPT-6 Sol and Luna run in Work and Codex, not in ordinary Chat.

How the limits behave

OpenAI publishes estimates, not counts: “These estimates are not fixed message limits.” What a message costs depends on the model, the size and complexity of the task, and whether it runs locally or in the cloud, where it may use more. Work and Codex draw on the same allowance, and weekly limits may apply on top of the five-hour window. Fast mode spends included usage at 2.5 times the standard rate, and Ultrafast at 8 times.

Pro removes the five-hour window, not the allowance. The Codex pricing page says Pro plans “currently have no five-hour limit”; OpenAI’s Pro help article does not mention a window and says that at a model’s limit it may be unavailable until its allowance resets, with ChatGPT showing the reset time when it can. On any plan, reaching a limit in the middle of a turn lets that turn finish. After that, Plus and Pro can buy credits, Business workspaces can buy shared workspace credits, and anyone can keep working locally with an API key at API rates. The usage dashboard, or /status in the Codex CLI, shows the live limit and reset time.

What this means if you run an agent

OpenClaw can sign in with a ChatGPT account and draw on these same allowances, so the answer to “what does the model cost?” can be “nothing beyond the subscription you already pay.” Four things decide whether that holds for an always-on agent:

• The model the allowance is spent on. A fresh OpenClaw setup selects GPT-6 Astra, the model with the smallest allowance on Plus. There, Luna buys roughly seventy times Astra’s messages, which is why it is the realistic choice for an agent that works all day; the model guide has the full comparison.
• Whether your OpenClaw can use GPT-6.1 Sol yet. In OpenClaw it is openai/gpt-6.1-sol. With an OpenAI API key it works from OpenClaw 2026.9.8, released October 3, 2026; earlier releases have no entry for the id. With a ChatGPT sign-in it takes 2026.9.9, released October 8: that release’s notes add the model to the Codex model list and move the bundled Codex to 0.160.0. 2026.9.8 bundled 0.158.0, and a user report shows ChatGPT refusing GPT-6.1 Sol requests from that version. Your account still has to offer it: until openclaw models list shows it as available, a ChatGPT sign-in reaches GPT-6 Sol and Luna, which Plus, Pro and Business still include: openclaw models set openai/gpt-6-luna. Updating does not switch the model for you.
• Whether Free or Go works on a server. OpenAI’s pages differ as of October 22, 2026. Its help article on using Codex with your ChatGPT plan says “Codex is included across ChatGPT plans, including Free and Go,” lists the Codex CLI among the apps you sign in with, and withholds only Codex Cloud from those two plans. Its pricing page names only the desktop app on the Free and Go cards. Check the pricing page before you plan an agent on a server around either plan. Plus is the cheapest plan both pages put in the CLI, and the cheapest with published usage estimates, so it is the realistic floor for an agent that works every day. A lapsed subscription drops to Free, and an agent set to a model Free does not include stops being served until you renew or switch models.
• The limits reset on a schedule. On Plus or a Standard Business seat, a busy afternoon can use up the five-hour window and pause the agent until it refreshes. Pro and Premium seats remove that window, but their allowance still runs out and resets. Configure a fallback model if silence isn’t acceptable.

Connecting from a server also needs one security setting enabled first. For the full cost picture including the machine the agent lives on, our OpenClaw cost breakdown is here.