You're deep in a refactor, Codex says "You've hit your usage limit", and the upgrade page is staring at you. Before you spend $100 or $200 a month, here's what the Pro tiers actually change for Codex — and how to tell whether you need them at all.
Short answer: if you hit Plus's Codex caps more than once a week, the $100 Pro tier usually pays for itself in un-blocked working hours. If you exhaust limits daily, run parallel agent tasks, or need the 1M-token context window, go straight to $200. If you only bump into limits during occasional crunch weeks, buying credits on Plus is cheaper than upgrading. The rest of this article is the reasoning.
What each plan actually gives you
Since April 9, 2026, ChatGPT Pro has been split into two tiers, sitting above the $20 Plus plan:
| Plan | Price (official) | Codex usage vs. Plus | Extras |
|---|---|---|---|
| Plus | $20/mo | 1x baseline | — |
| Pro 5X | $100/mo | 5x | Same model suite as Pro 20X, incl. GPT-5.5 Pro; Codex-Spark preview |
| Pro 20X | $200/mo | 20x | Everything in 5X, plus 250 deep research runs/mo and 1M-token context |
Two things worth underlining:
- The difference between the two Pro tiers is volume, not capability. Both get the same models. You can also switch between 5X and 20X anytime from Settings → My Plan, so picking "wrong" isn't a locked-in mistake.
- The launch promotion that gave the $100 tier 10x usage ended May 31, 2026. Plan around the standard 5x now.
How Codex limits actually work in 2026
Every plan carries two limits at once: a 5-hour rolling window and a weekly cap. Hitting either one stops you. Metering has been token-based since April 2026, so one "message" that refactors a large repo can cost tens of times more than a quick script fix. Local CLI messages and cloud tasks (including jobs kicked off from your phone) draw from the same pool.
As rough orientation, community-measured ranges per 5-hour window on GPT-5.5 look like this (workload-dependent, as of publication):
- Plus: ~15–80 local messages
- Pro 5X: ~75–400
- Pro 20X: ~300–1,600
Run /status inside Codex CLI (or check Codex settings → Usage) to see your own live numbers and reset times. And before concluding your plan is too small: in late June 2026 OpenAI's status page tracked an incident where Codex limits depleted faster than expected — occasionally the burn rate really is their bug, not your usage. A deeper troubleshooting walkthrough (in Chinese) is here: .