{"slug": "chatgpt-has-been-silently-routing-pro-to-mini-for-several-months", "title": "ChatGPT has been silently routing Pro to mini for several months", "summary": "OpenAI's $200-per-month ChatGPT Pro plan has been silently routing some Pro model sessions to the mini model for several months, according to a user who analyzed chat metadata and contacted support for two months without a resolution. The affected sessions showed no visual cue, and the model identified itself as GPT-5 mini, with metadata properties like 'i-cot' and 'gpt-5-5-mini' indicating the downgrade. The user publicly asks OpenAI to explain the issue, citing a lack of transparency for paying subscribers.", "body_md": "# ChatGPT has been quietly downgrading Pro model into mini. No response.\n\n## TL;DR\n\nUsing $200/month OpenAI's ChatGPT Pro(not lite), ChatGPT\nhas an issue where some Pro model sessions responded\nlike a mini model, or were routed into one, even when\nPro was selected in model picker. This is not normal,\nnever been documented, and has no visual cue whether\nit's affected or not. My finding relies on analyzing\n\"frozen/static\" chat metadata content\n(`resolved_model_slug`\n\n) rather than\npotentially hallucinated model outputs or chat\nresponses.\n\nI contacted support for two months[[img1]](#fnimg1)\nbut received only general/simple responses. Users paying\nfor the $200 plan expect research-grade intelligence,\nnot a same model offered to unpaid users. I publicly ask\nOpenAI to explain this issue.\n\nThis happened next day after I subscribed to the\n$200/month ChatGPT Pro plan[[1]](#fn1). While working as usual, the responses felt different.\nI can understand OpenAI's model performance can be\ninconsistent, but this change was too extreme to ignore.\n\nThe model's outputs fell far short of an expert or researcher quality, with the reasoning and thinking time significantly faster than normal (by dozens of times). Consequently, the performance was significantly below what is typically expected from the Pro model.\n\nAfter noticing this, I asked, \"What model are you?\" It identified itself as a completely different model: \"GPT-5 mini.\"\n\nAt first, I thought this was a model hallucination. Since there was no technical report for the Pro model, I thought its internal system might use mini models in some cases.\n\nHowever, my suspicion increased after comparing different clients logged into the same account. Some clients that had remained logged in for a long time(still not exact, some clients were affected too) still behaved as expected from the Pro model. New logins and other clients showed invalid behavior. This must not simply a response varied with prompt complexity (like simple prompt goes to a simpler model). Both the web and Android clients were affected just in case.\n\nThe affected Pro sessions showed the following behavior:\n\n- The Thinking UI did not appear, or reasoning finished almost instantly. I never saw text such as \"Thought for ...s\" in the suspected chats.\n- The model always identified itself as mini when asked.\n- I haven't reached any Pro model usage limit. When the Pro quota is exhausted, the ChatGPT UI disables the Pro option. It did not happen here at that moment.\n\nI then investigated ChatGPT data exports and backend\nnetwork responses. Based on an early reply from customer\nsupport, every chat in which the assistant identified\nitself as GPT-5 mini contained `i-cot`\n\nin the\n`resolved_model_slug`\n\nmetadata property.\nAbout a month later, another colleague researcher\nreported a similar case. In that case, that property was\nset to `gpt-5-5-mini`\n\n.\n\nWhile investigating that report, I also found a growing\nnumber of similar issue reports on social media\nrecently[[reports]](#fnreports). Thus I decided to write about this issue.\n\nCustomer support made this issue even more frustrating.\nI contacted support for about two months[[img1]](#fnimg1). The replies focused on general explanations, such as\nmodel hallucinations or clearing the browser cache. I\nnever received any explainer, actual investigation\nresults, or clear solution. Attempts to contact relevant\nOpenAI employees through X also received no response[[img2]](#fnimg2).\n\nWhat I consider most important about this issue is\n**transparency**. People pay $200 per month\nnot only for Codex, but also for research-grade\nreasoning. Yet this behavior continued without any clear\nnotice in the UI or by email. There was no visible\nindication/cue that something had changed. The issue\nmight not be limited to Pro users; Pro Lite and Business\naccounts have fewer quotas. Users have no reliable way\nto confirm whether they are receiving the service they\npaid for. I was affected for several days. For that\nperiod, the money was effectively wasted away.\n\nAI products are no longer toys nowadays. They are\nbecoming essential tools for research, development,\nwriting, analysis, and decision-making. Users of\nexpensive services expect more than better answers; they\nalso expect predictable service like usual. When a\nmodel's behavior suddenly changes and performance\nbecomes worse, the issue is no longer just performance.\nIt becomes a matter of **trust**.\n\nI am not claiming that these observations prove every\npart of my point.\n**However, OpenAI has provided no explanation at\nall.**\n\nUsers have every right to question whether they are receiving the service they paid for. When a provider stays silent or repeats useless answers, users lose trust in that provider.\n\nI'm not asking OpenAI to disclose confidential internal\ninformation. If this was a bug, respond that it was a\nbug. If it was part of an abuse-prevention or rate-limit\npolicy, explain that effectively. If fallback routing\nexists, explain when it occurs, address how it appears\nin the UI, and explain what level of service is\nguaranteed to paying users.\n__Customer support never gave me any information\nabout any of this.__\n\nSince\n__this article is solely focused for\nscientists/researchers/developers__, I still have a lot of stories to talk about. So I\nmight publish more articles when they do not address\nthis. Thank you to everyone who read this long and low\neffort post.\n\n## About Me\n\nComputer engineering undergraduate student in South Korea. I have some LLM-like knowledge through discussions with NLP doctoral student, AI (mostly vision, but work for all domains) researcher who studied neuroscience on their own, and industry professionals. I do research on computer graphics, desktop Linux, and software engineering.\n\n## How to Check Whether You Were Affected\n\nThe following ChatGPT shared chat sample links are available:\n\n-\nNormal/Expected behavior:\n[https://chatgpt.com/share/6a604011-8db0-83e8-b3bb-9399a4460919](https://chatgpt.com/share/6a604011-8db0-83e8-b3bb-9399a4460919) -\n5.6 Pro → 5-5-mini:\n[https://chatgpt.com/share/6a616b12-d9b8-83ee-b1e0-acb29487fe45](https://chatgpt.com/share/6a616b12-d9b8-83ee-b1e0-acb29487fe45) -\n5.5 Pro → i-cot:\n[https://chatgpt.com/share/6a603bf3-0590-83ee-9f56-6d8b573d43da](https://chatgpt.com/share/6a603bf3-0590-83ee-9f56-6d8b573d43da)\n\nAt the time of writing, there are two objective ways to check:\n\n- For chats you own, inspect the backend data in JSON via DevTools. If the chat was shared by someone else, fork it first. You will own that fork.\n- Inspect the server-side rendered/injected data of a shared chat. ChatGPT injects the shared-chat data (frontend state) into the HTML. This method is more complex because the data is not provided as normal JSON, but it's a reliable way to retrieve it when the first method is unavailable.\n\nIn both cases, check the prop value near\n`resolved_model_slug`\n\n. Expects\n`i-cot`\n\nor `gpt-5-5-mini`\n\nor\nsimilar. The first method is strongly recommended. The\nonly advantage of the second method is that it does not\nrequire login.\n\nWhile it was possible before, requesting a data dump\nexport and inspecting\n`resolved_model_slug`\n\nis unavailable now. The\nproperty seems to have been removed from exports\nrecently. A comparison is available in\n[\nthis GitHub Gist link ](https://gist.github.com/ilsubyeega/c76617cd97a071a4ef4dab81f2d0cc76).\n\n[[1]](#r1)\nChatGPT Plan | Pro. https://chatgpt.com/plans/pro/\nvisited 2026-07-30 (archived https://archive.md/H0nR2)\n\n[[reports]](#rreports)\n\nDisclaimer: Some portion of texts are not exact; some may truncated, machine-translated etc.\n\n@paulscu1 (06-14): Just tried OpenAI’s web UI, I’m\nshocked at just how bad the model routing is. I’m\npaying $100 a month, and I can’t get more than one\nreliable response from 5.5 pro before I start\ngetting routed to like 5.3 mini. It’s absurd product\ndesign, I will pay more, just please please fix this\nit doesn’t work\n[X Post](https://x.com/paulscu1/status/2065886816238985490)\n\n@Mortid_X (06-18): ChatGpt Pro extended 5.5 >\nGPT-5.3-mini model. This is some kind of bug. On my\nphone, everything works fine for me. I hope the\n@OpenAI team fixes this bug. Watch the video for\nproof: [video]\n[X Post](https://x.com/Mortid_X/status/2067343110024085958)\n\n@ludicrouslytrue (07-21): My web chat silently falls\nback to 5.5 mini? I'm on the 20x Pro plan\n@thsottiaux I thought codex and web limits weren't\nshared, oh well. [video]\n[X Post](https://x.com/ludicrouslytrue/status/2079532482928718203)\n\n@sizumita (07-23): Why is it a 5.5-mini when it's a\nPro model?\n[X Post](https://x.com/sizumita/status/2080269517176144169)\n\n@_machi47 (07-24): across all of my GPT chats on the\nPro plan, when using GPT 5.6 Sol Pro, it routes to\nGPT 5.5 mini. What is happening @thsottiaux ???? no\nnotices of degredation on the app itself. shows pro\nlevel, but does not reason at all, jus instantly\nspews garbage?[X Post](https://x.com/_machi47/status/2080394309715849434)\n\n@q8jx_ (07-24): does your gpt 5.6 pro say it's gpt\n5.5 mini too?\n[X Post](https://x.com/q8jx_/status/2080417495765184514)\n\n@RealWilliamHsu (07-24): You guys doing great but\nChatGPT with 5.6 Pro returned replies using GPT\n5.5-mini, how's that a genuine reply to the users??\n[X Post](https://x.com/RealWilliamHsu/status/2080469030499172453)\n\n@kumataro_x (07-26): @OpenAI Support rejected my\nrefund request, saying it could not confirm that my\nPro requests were handled by GPT-5.5 mini. I now\nhave structured data returned by ChatGPT's own web\nbackend showing exactly why this requires an\nengineering review. I selected Pro. The conversation\ndata shows: `default_model_slug: gpt-5-6-pro` But\ntwo separate requests show: `resolved_model_slug:\ngpt-5-5-mini` This is machine-readable metadata—not\na guess based on response speed or AI\nself-identification. I obtained the JSON directly\nfrom the response payload in my browser's DevTools\nNetwork panel while loading the original ChatGPT\nconversation. The conversation ID in the page URL\nexactly matches the JSON. I also preserved the\noriginal file and its SHA-256 hash. OpenAI's Help\nCenter says Pro is powered by GPT-5.6 Sol Pro. It\ndocuments GPT-5.5 Instant mini as an Instant-limit\nfallback—not as a fallback for selected Pro\nrequests. @OpenAI, please explain what\n`resolved_model_slug` means and check the actual\nserving model for my two request IDs. If Pro served\nthem, why did your backend return mini? If mini\nserved them, acknowledge the routing issue and\nreconsider the refund.\n[X Post](https://x.com/kumataro_x/status/2081060343955296414)\n\n@whatfordinner12 (07-28): I pay $100/month for\nChatGPT Pro. The UI showed GPT-5.6 Pro. Pro quota\nwas charged. But the request was routed to\nGPT-5.5-mini. But after I submitted screenshots,\nchat links, and quota evidence, Support twice\ndismissed it as “AI hallucination.” (1/5)\n[X Post](https://x.com/whatfordinner12/status/2082056790091743508)\n\n@TuringMeme (07-28): I’m a ChatGPT Pro user. Today I\nnoticed that selecting GPT-5.5 Pro automatically\nswitched my conversations to GPT-5.5 Mini. Support\nexplained that Mini is the fallback after reaching\nthe Pro usage limit. I understand having a fallback\nmodel, but the problem is: the quality difference is\nobvious, and there is no clear indication of when\nPro access will be restored. Please show the limit\nstatus and reset time. Transparency matters for Pro\nusers.\n[X Post](https://x.com/TuringMeme/statu</p>s/2082103025964023906)\n\n@zhihe_md (07-29): I am extremely dissatisfied with\nChatGPT’s current risk-control and model-routing\nsystem. I subscribe to the 20× Pro plan, yet I\npreviously used fewer than 2 pro requests per day on\naverage, and my Codex usage rarely exceeded 30% of\nthe available quota. My overall usage has been\nmodest, with no apparent indication of abuse or\nabnormal activity. Despite this, all of my GPT-5.6\nSol Pro requests are now being routed to GPT-5.5\nMini. In practice, this means that the main\nhigh-performance model included in my subscription\nis no longer available to me. The system has\nprovided no clear explanation, no estimated duration\nfor the restriction, and no meaningful appeal or\nreview process. I paid for a higher-tier Pro\nsubscription, but my access has been silently\ndowngraded despite normal and limited usage. This\nopaque and indiscriminate risk-control mechanism\nseriously undermines the value of the subscription\nand my trust in the advertised Pro benefits. Please\nconduct a manual review of my account, explain the\nspecific reason for this restriction, and restore\naccess to the models and service level included with\nmy subscription.\n[X Post](https://x.com/zhihe_md/status/2082272498549506488)\n\n\"5.6 pro model has been automatically downgraded and\nrouted to the 5.5 mini model since its release\"\n(07-23, 9 users)\n[OpenAI Developer Community Post](https://community.openai.com/t/5-6-pro-model-has-been-automatically-downgraded-and-routed-to-the-5-5-mini-model-since-its-release/1387941)\n\n\"Pro subscription routing to GPT-5.3 Mini for 48+\nhours — no reset timer visible\" (06-17, 32 users)\n[OpenAI Developer Community Post](https://community.openai.com/t/pro-subscription-routing-to-gpt-5-3-mini-for-48-hours-no-reset-timer-visible-case-10179682/1383983)\n\nu/ilsubyeega (06-22): P.S.: The Pro model quietly\nredirects to GPT-mini based model. $200/month\n[Reddit Post](https://www.reddit.com/r/ChatGPTPro/comments/1v34sg1/ps_the_pro_model_quietly_redirects_to_gptmini/)\n\nu/immortalsol (06-18): I am being silently rerouted\nto GPT-5.3 mini when using Pro Extended mode on\nChatGPT webapp. What the hel lis going on? only\nstarted happening today.\n[Reddit Post](https://www.reddit.com/r/OpenAI/comments/1u8wyfq/i_am_being_silently_rerouted_to_gpt53_mini_when/)\n\nu/TrainingEngine1 (05-24): My 20x Pro Plan has every\nmessage reverted to 5-Mini and OpenAI refuses to\nhelp. [content deleted]\n[Reddit Post](https://www.reddit.com/r/ChatGPTPro/comments/1tm4sq5/my_20x_pro_plan_has_every_message_reverted_to/)\n\nu/Pleasant-Contact-556 (2025-05-02): What are i-cot\nand i-mini-m? I got rate-limited on my pro\nsubscription. It happens occasionally for who knows\nwhat reason, and when it does you can tell because\nall of the CoT models route to something.. lesser..\n[Reddit Post](https://www.reddit.com/r/OpenAI/comments/1kcrvra/what_are_icot_and_iminim/)", "url": "https://wpnews.pro/news/chatgpt-has-been-silently-routing-pro-to-mini-for-several-months", "canonical_source": "https://articles.fmt.kr/2026-07/oai", "published_at": "2026-07-31 06:07:05+00:00", "updated_at": "2026-07-31 06:22:17.245799+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-products", "ai-ethics"], "entities": ["OpenAI", "ChatGPT Pro", "GPT-5 mini"], "alternates": {"html": "https://wpnews.pro/news/chatgpt-has-been-silently-routing-pro-to-mini-for-several-months", "markdown": "https://wpnews.pro/news/chatgpt-has-been-silently-routing-pro-to-mini-for-several-months.md", "text": "https://wpnews.pro/news/chatgpt-has-been-silently-routing-pro-to-mini-for-several-months.txt", "jsonld": "https://wpnews.pro/news/chatgpt-has-been-silently-routing-pro-to-mini-for-several-months.jsonld"}}