Codex · AI costs · 2026-10-17

ChatGPT Pro 200 limits drop after 29 Oct: how to make Codex usage last

If you've been on ChatGPT Pro 200 since September, your Codex allowance is about to get smaller. The price stays the same. This post sets out what OpenAI's help pages actually say about the change, and then covers the documented ways to make a smaller allowance last: picking the right model and speed, keeping context small, and moving heavy repeat work to the API with a spend cap.

This guide pairs with Spend Caps Kit, and if you want the next step, read Prompt caching not working? Check it in OpenAI and Anthropic (Next.js).

Everything below comes from OpenAI's help centre and Codex docs, checked on 17 October 2026. Plans and limits change often, so check the pages linked here before you make a decision. We aren't affiliated with or endorsed by OpenAI; ChatGPT and Codex are OpenAI's trademarks, used here only to name its products.

What changes, and who it affects

From the About ChatGPT Pro tiers help page:

There are three Pro plans: Pro 100 ($100 a month), Pro 200 ($200) and Pro 500 ($500). "Pro 200 includes more usage than Pro 100. Pro 500 offers the highest included usage of the three plans."
New Pro 200 subscriptions get a lower allowance. "New subscriptions that aren't eligible for grandfathering include a lower usage allowance than previously offered with Pro 200 to reflect our increasingly efficient models. The monthly price remains $200."
Existing subscribers keep the old allowance for now. "Grandfathering applies if you had an active Pro 200 subscription at any point from September 22, 2026 through 10 a.m. Pacific Time on September 29, 2026. Eligible customers can use the previous included allowance through October 29, 2026 while their Pro 200 subscription is active."
After that, they move down too. "After that date, your subscription will move to the lower included usage allowance. Your subscription price stays at $200/month. Only users affected by this change will receive an email with additional details."

So only grandfathered Pro 200 subscribers are affected by the 29 October date. The help page describes this change only for Pro 200; it doesn't say Pro 100 or Pro 500 allowances are changing. And anyone who took out Pro 200 after the window already has the lower allowance.

How much lower? OpenAI's help pages don't give a multiplier. According to text posted on Hacker News as OpenAI's message to affected subscribers, "your included usage in ChatGPT Work and Codex will decrease from 20 times to 10 times the ChatGPT Plus allowance", starting 30 October 2026. We haven't seen the original message ourselves. The safest figure is the one on your own usage dashboard after the change.

What uses up your allowance

The same help page says: "Usage depends on the model, the amount of context processed, reasoning, tool use, information retrieval, caching, and speed mode." The Codex pricing page adds that "larger projects, long-running tasks, or extended sessions that require the agent to hold more context will use significantly more per message", and that "prompt length alone isn't a reliable estimate".

It also says local messages and cloud chats share one allowance, and "Cloud tasks may use more of your allowance than local messages." ChatGPT Work and Codex share usage too, so a long Work session draws on the same allowance as your coding.

Each of the habits below works on one of those factors.

1. Check where you stand first

The Codex pricing page points to a usage dashboard for current limits and reset times. In the Codex CLI, the slash command docs describe /usage as "View account token usage or use a rate-limit reset" and /status as "Display session configuration and token usage". The pricing page also says you can use /status to see your remaining limits during a session, so if one command doesn't show what you need, try the other. OpenAI suggests checking the dashboard "every week or two to understand your pace and remaining capacity."

Do this in the week before 29 October and again in the week after. Then you'll know whether the change matters for how you actually work, instead of guessing from a multiplier.

2. Turn off Fast mode unless you need it

Speed mode is the factor with the clearest numbers. From the help page: "Compared with Standard mode for the same model, Fast mode uses your included allowance at 2.5 times the rate, and Ultrafast for GPT-6 Astra and GPT-6.1 Sol uses it at 8 times the rate." Ultrafast is only on Pro 500, so on Pro 200 the one to watch is Fast.

The Speed docs say /fast toggles it in the CLI, and /statusline can show it in the footer so you notice when it's on. If you've set it as your default in config.toml, remove service_tier = "fast" there. At 2.5 times the rate, the same work in Fast mode takes 2.5 times as much allowance as it would in Standard.

3. Match the model and reasoning effort to the task

The Models page recommends GPT-6.1 Sol "for complex coding and agentic workflows" and says to "Use Luna for focused, repeatable tasks." On reasoning effort, it says higher effort "takes longer and uses more tokens. Start with the default effort and increase it when" a task needs it.

Codex's pricing table shows how wide the gap is. For Plus, it estimates 15–160 local messages per five hours with GPT-6.1 Sol and 350–3,000 with GPT-6 Luna. Those are Plus estimates, not Pro figures, and OpenAI says "These estimates are not fixed message limits". The ratio is still a useful guide.

You can switch per session with /model, or set a lighter default in ~/.codex/config.toml and move up only when you need to:

# ~/.codex/config.toml
model = "gpt-6-luna"
model_reasoning_effort = "high"

Both keys are in the Codex config reference. The available models and effort levels depend on your plan and client, so check /model to see what you can pick. The Models page says "Start with High for Luna", so the example uses high.

4. Keep context small

Context is the factor that's easiest to let grow without noticing. The pricing page lists tips for making limits last longer:

"Control the size of your prompts. Be precise with the instructions you give the agent, but remove unnecessary context."
"Limit source material. Provide only relevant files and, when possible, narrow the sources or date range."
"Match the output to the need. Define the audience, format, and length, and separate required work from optional improvements."
Reduce the size of your AGENTS.md, for example by nesting files within the repository.
"Limit the number of MCP servers you use. Every MCP server adds more context to your messages and uses more of your limit."

Two CLI commands help with long sessions. According to the slash command docs, /compact "replaces earlier turns with a concise summary, freeing context while keeping critical details", and /new starts a fresh chat in the same CLI session. If a thread has drifted onto a different task, a new chat avoids carrying the old context into every message.

On AGENTS.md: the AGENTS.md guide explains that Codex reads the files from the project root down to your working directory and joins them, up to project_doc_max_bytes (32 KiB by default). Every message pays for that text, so keep the root file short and put team- or folder-specific rules in nested files, where they only load when you work in that part of the repo.

5. Plan around the reset, not against it

The pricing page says that if you reach your limit during an active turn, "the agent will be able to continue working on that turn, subject to fair use limits." It also notes that Pro plans "currently have no five-hour limit", though weekly limits may apply.

In practice, start big tasks when you have room, not when the dashboard is nearly empty. Batch related changes into one well-scoped task rather than many small back-and-forth turns, each of which resends context. That last point is our inference from the "context" factor above, not a figure OpenAI publishes.

6. Move heavy repeat work to the API, with a cap

Some work doesn't need a subscription allowance at all: a nightly job that summarises logs, a CI step that writes release notes, or a batch of files to classify. OpenAI documents three routes:

Codex with an API key. The auth docs say Codex can sign in with an API key, and that "OpenAI bills API key usage through your OpenAI Platform account at standard API rates." The pricing page confirms that "All users may also run extra local chats using an API key". Cloud features such as GitHub code review aren't available this way.
Non-interactive runs. The non-interactive mode guide uses codex exec for "CI, pre-merge checks, scheduled jobs".
The Batch API, for work that doesn't need Codex. If the job is plain model calls, the Batch guide offers "50% lower costs" with "a clear 24-hour turnaround time", for jobs that don't need an immediate answer.

A scheduled job with an API key might look like this. The commands are from the auth and non-interactive docs:

# On a machine you control: sign Codex in with an API key (billed at API rates)
printenv OPENAI_API_KEY | codex login --with-api-key

# In a scheduled job or CI step
codex exec "generate release notes for the last 10 commits" | tee release-notes.md

In CI, don't expose the key to the whole job. OpenAI's non-interactive guide suggests the Codex GitHub Action on GitHub, or setting CODEX_API_KEY only on the codex exec step.

Moving work to the API swaps a fixed allowance for a bill with no ceiling, unless you set one. Do that first:

Use a separate project and key for the job, and set a monthly hard limit on that project. The spend limits guide says requests then fail with a 429 and project_spend_limit_exceeded, and warns that "Enforcement is not instantaneous, so recorded spend can slightly exceed the configured amount."
Add spend alerts below the hard limit, so you hear about a runaway job before it stops.
Keep the stable part of the prompt at the start. The prompt caching guide says caching is on by default for supported models and reuses a matching prefix, "the unchanged tokens at the beginning of a prompt". Cached input is billed at a lower rate on most models; see the API pricing page.

Check API pricing for the model you plan to use before you move anything. Whether the API works out cheaper than staying within the subscription depends on your volume and model, and we haven't compared them for any specific workload.

Checklist before 29 October

Check whether you're grandfathered: an active Pro 200 subscription at any point from 22 September to 10 a.m. PT on 29 September 2026.
Look at your usage dashboard now, and again a week after the change.
Turn Fast mode off by default, and use it only when speed matters.
Default to a lighter model and effort, and move up per task.
Trim AGENTS.md and disable MCP servers you aren't using.
Use /compact or /new when a session gets long or changes topic.
Move scheduled and CI work to an API key on a separate project with a hard spend limit.

Codex credits explained: this post covers your included usage limits and how to track them; that one covers credits, which Codex uses after your included usage runs out, plus earned resets and capping auto reload.
OpenAI and Anthropic spend limits per user in Next.js: if the repeat work becomes a feature in your own app, this is how to cap spend per user, not just per project.

Disclosure: this is our own paid product. If the work you move to the API becomes part of your own Next.js app, calling OpenAI or Anthropic on your users' behalf, the Spend Caps Kit ($19 plus VAT where applicable) adds per-user rate limits and a monthly dollar cap per user, with one Supabase migration, a price table for 16 OpenAI and Anthropic models, and a provider-side backstop checklist. It doesn't change your ChatGPT or Codex plan limits, and it doesn't replace a project spend limit.

Sources (checked 17 October 2026):