You pay $200 a month for ChatGPT Pro. You pick the Pro model in the dropdown. What if I told you some of your sessions were quietly running on the free mini model this whole time? ChatGPT Pro routing to mini is a real issue, and OpenAI hasn’t said a word about it.
What is actually happening?
A detailed investigation published in July 2026 laid out the problem. ChatGPT Pro subscribers, the people paying $200 a month for access to OpenAI’s most capable models, have been getting responses from GPT-5 mini in some sessions. Not the model they selected. Not the model they’re paying for. The cheaper, less capable version that free users get.
And there’s no warning. No indicator in the interface. No message saying the session was downgraded. The dropdown still says Pro. You type your prompt expecting top-tier reasoning, and you get something noticeably worse. This ChatGPT Pro routing to mini problem has no visible symptoms in the UI.
The person who uncovered this, who goes by @ilsubyeega online, spent two months documenting the issue. They contacted OpenAI support multiple times. The responses were generic and unhelpful. No acknowledgment, no fix, no explanation.
How someone figured this out
Here’s where it gets technically interesting, even if you’re not a developer.
Every ChatGPT response carries hidden metadata. One field, called resolved_model_slug, records which model actually generated the response. Not which model you selected, but which one actually ran. The investigator downloaded their chat data exports and checked this field across dozens of conversations. You can read the full technical investigation here.
The results were damning. In sessions where the Pro model felt “off,” the metadata consistently showed “i-cot,” a marker associated with the mini model. Meanwhile, sessions that performed as expected showed the Pro model identifier. This ChatGPT Pro routing to mini pattern was not random.
They went further. They logged into ChatGPT on multiple browsers and devices using the same Pro account. Chrome showed the downgrade. Firefox sometimes didn’t. Fresh logins were more likely affected than sessions that had stayed logged in for a while.
Then they tried asking the model directly: “What model are you?” In affected sessions, the model responded “GPT-5 mini.” In normal sessions, it correctly identified itself as the Pro model.
Signs your Pro session might be affected
You won’t see a popup or a warning. But there are patterns that hint something’s wrong.
The most obvious sign is reasoning speed. ChatGPT Pro is supposed to “think” before responding. You see the thinking indicator, the “Thought for X seconds” message. When the session is downgraded to mini, that thinking phase either doesn’t appear or finishes almost instantly. The response feels rushed.
Quality drops noticeably too. Pro handles complex, multi-step reasoning better than any model in OpenAI’s lineup. Mini is designed for speed, not depth. If you’re asking a nuanced question about code architecture, legal analysis, or strategic planning, and the response feels superficial, that’s a red flag.
Another tell: consistency across the same prompt. If you ask the same complex question in two different sessions and get wildly different quality levels, one of them might be running on mini. Pro should be consistently excellent. Mini will be consistently average.
The evidence from ChatGPT metadata
The metadata trail is the strongest evidence we have. Here’s why it matters.
When you request your ChatGPT data export, you get a file containing every conversation. Each conversation includes technical metadata that most users never look at. The resolved_model_slug field is the ground truth. It records exactly which model processed your request, regardless of what the UI showed.
The investigator found that the pattern was consistent. Affected sessions showed the mini model identifier. Normal sessions showed Pro. There was no ambiguity.
This matters because model self-identification can be unreliable. Language models sometimes hallucinate their own identity. But metadata doesn’t lie. It’s a server-side record of what actually ran.
What OpenAI has said (or not said)
Nothing. That’s the short answer.
The investigator contacted support repeatedly over two months. The responses were generic, the kind of copy-paste replies you get from tier-one customer service. There was no acknowledgment of the ChatGPT Pro routing to mini issue. No explanation of whether it’s intentional behavior or a bug. And certainly no timeline for a fix.
To be fair, there could be a legitimate technical explanation. Maybe OpenAI’s infrastructure dynamically routes requests based on server load. Maybe there’s a load-balancing system that occasionally sends Pro requests to mini during peak times. These things happen at scale.
But here’s the problem with that defense: none of it is documented. OpenAI’s Pro page promises access to their most capable models. If they’re silently routing to mini under certain conditions, users deserve to know when and why. Charging $200 a month for premium access while quietly serving the free model is, at minimum, a transparency failure.
How to check if it is happening to you
You don’t need to be a developer to investigate this.
The simplest method is to ask. Open ChatGPT with Pro selected and ask “What model are you?” Do this across multiple new chats. If some sessions identify as mini while others say Pro, you might be affected.
This isn’t foolproof since models can hallucinate their identity. But if you’re consistently getting “mini” responses on a Pro plan, something’s wrong.
For a more reliable check, request your data export. OpenAI lets you download all your chat data through Settings > Data Controls > Export. Once you have the file, search for resolved_model_slug in the data. If you see “mini” or “i-cot” values in conversations where you selected Pro, that confirms the routing issue.
Finally, pay attention to response quality. If you’ve been using Pro for a while and certain sessions feel suspiciously weak, they might not have been running Pro at all. Cross-reference with the metadata if you can.
Is your Pro subscription still worth it?
ChatGPT Pro routing to mini doesn’t mean the entire subscription is a scam. When Pro works correctly, it’s still the best model OpenAI offers for complex reasoning tasks. Most sessions are probably fine. But the fact that some aren’t, combined with OpenAI’s silence, is reason for caution.
If you rely on Pro for work, you should be monitoring your outputs. The whole point of paying $200 a month is consistency. You need to know that every session delivers the intelligence you’re paying for. If you can’t trust the model selector, the value proposition gets shaky.
For users who only need Pro occasionally, this issue might not matter much. But for power users who depend on frontier-level reasoning every day, it’s worth paying attention to. Check your metadata. Monitor your response quality. And if you’re consistently seeing signs of downgrading, consider whether that $200 is better spent elsewhere.
The broader lesson here applies beyond ChatGPT. When you pay for a premium AI service, verify you’re getting what you paid for. Different models serve different purposes, and the gap between them can be enormous. Trusting the UI to tell you the truth isn’t enough anymore.
