{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreihyhbsg5evbxy2fozmgdfividlqgwlduouidon4bncnljdaqzyw5e",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mm4g2dapz3g2"
  },
  "path": "/t/paying-for-chatgpt-plus-just-to-be-locked-out-while-5h-remaining-still-shows-feels-misleading/1381028?page=3#post_45",
  "publishedAt": "2026-05-18T06:10:54.000Z",
  "site": "https://community.openai.com",
  "tags": [
    "Zamin"
  ],
  "textContent": "Zamin, don’t worry.  **I’m ok!**\nThe core problem in one sentence: Codex charges in tokens but shows limits in hours. Two completely different units that are never openly converted.\n#1 **— Official OpenAI documentation.**\nSince April 2, 2026, pricing officially moved to token-based billing: credits per million input tokens, cached input tokens, and output tokens. Source: OpenAI Help Center, Codex Rate Card.\n#2 **— OpenAI admits it themselves.**\nOfficially: larger codebases and longer sessions use significantly more per message. And system files burn the limit before a single line of code is written. Source: _OpenAI Help Center._\n#3 **— Real forum data.**\nA single prompt for a one-line change consumed 7% of the weekly limit and 25% of the 5-hour window. Source: _OpenAI Developer Community._\n#4 **— My personal case — disproportionate billing.**\nTask: make a commit. Execution time: 3 minutes 46 seconds. Deducted: 7% of the 5-hour window. Mathematically 3 minutes 46 seconds out of 5 hours is under 1% of the time. But 7% was charged. Because billing runs on tokens, not time. And how many tokens equal one percent — _**OpenAI shows this nowhere.**_\n#5 **— Codex lives in a parallel universe — separate problem.**\nI asked Codex how long today’s task took. It answered: **35-50** minutes. Real clock time: under **10** minutes. It estimates time by _token_ volume, not real minutes. Meaning Codex itself doesn’t know how much real time has passed​.\n_**Full picture:**_ the interface shows hours, billing runs on tokens, Codex estimates time in tokens too. Three different measurements. None match real time. The user pays for all three simultaneously without understanding any of them.\n_For this reason, most likely, the panel shows a bug that there is still time, but in reality, the tokens have run out._ Or maybe I’m wrong.\n\n_**And one more thing — a concrete suggestion.**_\nThe fix is not complicated. Instead of showing percentages that mean nothing to the user, show a real-time token counter next to the clock.\nNot “**68%** ” — but “**45,000** tokens remaining.”\nAfter each task — a receipt: how many tokens were spent, what’s left.\nBefore starting a task — an estimate: this will cost approximately X tokens.\nOne unified timeline — not two separate counters (**5h and Weekly**) that block each other silently without warning.\nRight now the interface shows time, charges tokens, and the two counters run independently without telling the user which one is about to run out first. That’s not a design choice. That’s a blind spot.\n_And yes — looking at the screenshot: 68% remaining on the 5h window, 0% on Weekly, system blocked. The interface never warned that Weekly would run out first. That is exactly the bug._",
  "title": "Paying for ChatGPT Plus Just to Be Locked Out While “5h Remaining” Still Shows Feels Misleading"
}