# Codex Limits: ChatGPT Plus vs Pro vs Business

Every ChatGPT plan includes Codex; the window sizes differ. How Plus, Pro, and Business capacity compares, which tier fits which workload, and the cost math as estimates.

*Published 2026-06-12 · https://proxyllm.ai/blog/codex-limits-plus-vs-pro*

Every ChatGPT plan includes Codex; what you buy with a bigger plan is bigger windows. Plus covers a working developer's interactive use, the Pro tiers exist for agent workloads, and Business allocates capacity per seat for teams. OpenAI varies the exact limits by plan and model and adjusts them over time, so current figures live at developers.openai.com/codex/pricing; this page compares the tiers structurally and gives our planning estimates for what each absorbs. The mechanics of the windows themselves are in [Codex usage limits, explained](/blog/codex-usage-limits-explained).

## The comparison

All tiers share the same machinery: a rolling window of roughly five hours, weekly components on some plans, one pool shared across every Codex surface, and on-demand credits for Plus and Pro. The table below is how we size plans for customers. The dollar figures are our planning estimates of monthly API-equivalent throughput, derived from observed workloads; they are estimates, never guarantees.

| Plan      | Monthly price | Our planning estimate   | Credits available         |
| --------- | ------------- | ----------------------- | ------------------------- |
| Free / Go | $0 / entry    | Evaluation scale        | See OpenAI's pricing page |
| Plus      | $20           | ~$700 API-equivalent    | Yes                       |
| Pro 5x    | $100          | ~$3,500 API-equivalent  | Yes                       |
| Pro 20x   | $200          | ~$14,000 API-equivalent | Yes                       |
| Business  | Per seat      | Scales with seats       | See OpenAI's pricing page |

Edu and Enterprise mirror Business: Codex included, per-seat capacity, admin controls, pricing by agreement.

## Which tier fits which workload

**Free and Go** answer one question: is Codex any good for your work? Run it on a real repo for a week. The windows are small, and that is fine for the job they do.

**Plus** is the daily driver. Interactive coding, a few `codex exec` scripts, occasional cloud tasks. The 5-hour window will pinch on your heaviest days, and the weekly component is where enthusiastic agent experiments go to die. If you hit limits on Plus once a quarter, stay on Plus.

**Pro 5x** is where agent workloads start. Scheduled jobs, an automation or two running through the day, regular long sessions. As arithmetic: if your workload would cost around $3,500 a month at API rates, this tier is $100 and ProxyLLM is a flat $129 on top. A $3,500 monthly OpenAI API bill maps to those two numbers, whatever the volume. That sentence is the entire category, and it stays an estimate until your own request log confirms it.

**Pro 20x** is for fleets: parallel agents, always-on pipelines, agency-scale throughput. If you are asking whether Pro gives you API-style access to go with that capacity, the honest answer is its own article: [is there an API for ChatGPT Pro?](/blog/is-there-an-api-for-chatgpt-pro)

**Business** fits teams that want one invoice and admin control. Capacity scales by adding seats, and each seat is a person: OpenAI's terms do not allow sharing a login to pool capacity. Per-seat windows make Business strong for many moderate users and weak as a substitute for one very heavy account.

## The signal you are on the wrong tier

Watch which limit you hit and how often:

- Hitting the 5-hour window occasionally is normal on every tier; it is a pace limiter.
- Hitting the weekly cap monthly means your sustained load outgrew the plan. Upgrade, or add a lane.
- Buying [on-demand credits](/blog/codex-credits-explained) more than rarely means you are paying metered rates on top of a flat plan, which is the worst of both. The next tier up is usually cheaper than habitual credits.

The upgrade question is arithmetic, not loyalty: compare a month of credit purchases against the price gap to the next tier.

## When one account is the real bottleneck

Past a certain load, no single tier behaves. Burst days blow through windows that average days never touch, and a plan sized for your peaks is overpriced for your median. The fix is structural rather than tier shopping: run your sustained load on a right-sized plan and give overflow somewhere to go, a second account you own or your own OpenAI API key. We automate that ordering, with a per-request log of which lane served each call, and the design is documented in [fallback lanes, explained](/blog/codex-hosted-fallback-lanes-explained).

Plan capacity is the input; your workload is the test. Put your actual monthly API spend into the [calculator](/calculator) and it will say which tier covers it, with the math shown.