On the same $20/month Plus plan, Codex sometimes stalls after 20 minutes and sometimes runs all afternoon without hitting the limit. It is not luck: it is the model you picked for the session.
Why does Codex hit the limit so fast sometimes?
Because the quota is not counted in tokens, it is counted in messages, and the message count on the same plan varies up to 40x depending on the model. On Plus, GPT-6 Astra gives you 5 to 45 messages every 5 hours; GPT-5.6 Luna gives you 250 to 2,000 in the same window.
| Plan | GPT-6 Astra | GPT-5.6 Sol | GPT-5.6 Terra | GPT-5.6 Luna |
|---|---|---|---|---|
| Plus ($20/mo) | 5 to 45 | 10 to 100 | 25 to 200 | 250 to 2,000 |
| Pro 5x ($100/mo) | 25 to 225 | 50 to 500 | 125 to 1,000 | 1,250 to 10,000 |
| Pro 20x ($200/mo) | 100 to 900 | 200 to 2,000 | 500 to 4,000 | 5,000 to 40,000 |
| Business ($20/user/mo) | 5 to 45 | 10 to 100 | 25 to 200 | 250 to 2,000 |
Messages per rolling 5-hour window. Source: learn.chatgpt.com/docs/pricing, checked on 2026-09-05.
I got a bigger launch-time limit. Does that explain the drop?
Partly. Codex had a double-limit period at launch and that promotion has already ended, confirmed by official OpenAI community moderation. Whoever joined during that window feels the "drop" now, but it is not a new cut, it is the end of a temporary bonus.
On top of that, OpenAI itself acknowledged two technical factors that made part of the user base feel the quota run out faster in 2026: a Sol model inefficiency in July, with a fix targeting sessions roughly 18% longer, and a temporary drop in cache hit rate in August, which also speeds up consumption.
I hit the limit right now, what do I run, in what order?
Three commands, verified straight from the openai/codex source code, cover most cases before you have to wait for the reset.
/status: shows the active session's configuration and token usage, without leaving the terminal.codex login status: confirms how you logged in. ChatGPT subscription has a message cap; API key login bills by usage and skips that quota./model: switches the model and reasoning level inside the session.
In practice: switching to Terra or Luna is not a downgrade for every task. For code review, a simple refactor, or a quick question, a lighter model does the same job for a fraction of the Astra or Sol quota.
Still stuck after that? The usage window is rolling and resets every 5 hours, there is no other command that fixes it besides waiting.
Is there a way out of the "which model fits my plan" math?
The underlying problem is not which command to run, it is that each model has a different quota price inside the same fixed plan. On Verboo Code that table does not exist: plans have no per-model token cap.
| Plan | Price/mo | Per-model token cap |
|---|---|---|
| Junior | R$ 99 | none |
| Pro | R$ 249 | none |
| Max | R$ 499 | none |
| Ultra | R$ 1,299 | none |
Prices checked directly against Stripe on 2026-09-05.
If the question before every task is "will this model blow my quota", Verboo Code takes that question out of the equation: unlimited tokens on every plan, switching models becomes a quality call, not a quota gamble. Check it out at verboo.ai.
