How do you fix a slow ChatGPT?
Check the thinking level first, then the thread, then OpenAI. OpenAI's own slow response guide covers cache, status, browser and network. It doesn't mention thinking levels or long threads, and those are the two causes you can see on your screen. Work down this list:
- Open the model menu and check the Thinking slider. Medium, High and Extra High reason before they answer (on Free and Go, it's the Think option). Switch to Instant for anything that needs no reasoning.
- Open a new chat and send one short prompt. If it answers fast, the long thread is the problem. Carry the work over with a handoff summary.
- Check status.openai.com. Read the Conversations row. A posted incident means you wait, because nothing on your side will fix it.
- Try a private window. It turns off most extensions, which OpenAI says "can interfere with web applications".
- Clear ChatGPT's site data, then sign back in. OpenAI says sudden slowness is "most often" outdated cache data.
- Turn off the VPN and try mobile data. If ChatGPT is fast there, your wifi, office network or VPN is the bottleneck.
| What you see | Likely cause | First fix |
|---|---|---|
| A "Thinking" status before any text appears | A Medium, High or Extra High thinking level | Switch to Instant |
| Typing and scrolling lag in one old chat | A long thread the browser has to draw | New chat with a handoff |
| Every chat is slow, on every device | An OpenAI incident or peak hour load | Check status.openai.com, then wait |
| Slow in one browser, fine in another | An extension or stale site data | Private window, then clear site data |
| Slow on office wifi, fine on mobile data | A VPN, proxy or company filter | Turn the VPN off, or ask IT |
| The reply stops halfway with an error | The connection dropped mid answer | The network error fixes below |
Why is ChatGPT so slow when it is thinking?
Thinking is time ChatGPT spends reasoning before it answers, and every step up the slider adds more of it. The Thinking slider runs from Instant through Medium, High and Extra High. When GPT-6 is selected, GPT-6 Sol powers all four, according to OpenAI's GPT-6 help page. OpenAI doesn't say which API effort setting each slider step uses. So we matched them by name only: Instant to the API's no reasoning setting, Medium to medium, High to high, and Extra High to xhigh. Artificial Analysis times each of those API settings:
| GPT-6 Sol API setting, as of October 2026 (ChatGPT slider name it matches) | First word of the answer | 500 token answer finished | Writing speed once it starts |
|---|---|---|---|
| No reasoning (Instant) | 0.8 seconds | 5.6 seconds | 104 tokens a second |
| medium (Medium) | 2.7 seconds | 11.9 seconds | 54 tokens a second |
| high (High) | 24.9 seconds | 29.5 seconds | 108 tokens a second |
| xhigh (Extra High) | 68.7 seconds | 73.2 seconds | 111 tokens a second |
Nearly all of the extra wait is thinking. Once text starts, xhigh writes as fast as the no reasoning setting (111 tokens a second against 104), so the gap sits before the first word. These are API numbers, not app numbers. Artificial Analysis sends a prompt of about 10,000 tokens and reports the median of the past 72 hours, per its methodology page. Nobody publishes a stopwatch for the ChatGPT app, so treat the table as the size of the gap, not the exact wait on your screen. OpenAI also says a GPT-6 reply in the app "may begin while ChatGPT continues thinking," so text can show up before the thinking ends.
Plus gets Medium and High. Extra High comes with Pro, Business and Enterprise. On Plus and Pro, OpenAI says Instant "doesn't automatically switch to a higher thinking level for complex requests," so a slow Instant reply isn't hidden thinking. Free and Go run GPT-5.6 Luna, which OpenAI calls "the fastest and lowest-cost model in the GPT-5.6 family". Their Think option uses Luna too.
OpenAI also tunes thinking time between releases. Its release notes say it "lowered the Standard and Light thinking time" for GPT-5.2 Thinking on 10 January 2026, then made another small cut on 3 February. So a ChatGPT that feels slower lately may be running a setting OpenAI moved. What is a reasoning model explains what that extra time buys.
Why does ChatGPT get slow in a long conversation?
A long ChatGPT conversation gets slow mostly in the browser, according to users on OpenAI's own forum. In an OpenAI forum thread opened in May 2024 and viewed 25,224 times, the first poster says chats were "getting slow after about 30 messages" in Chrome and Edge. The same chats ran fine in the mobile app. A June 2026 thread blames the interface, "not the model itself necessarily": scrolling, editing and loading.
OpenAI's own notes back the browser side. Its ChatGPT release notes for 21 August 2026 say long conversations now load faster on the web, because "ChatGPT loads messages in smaller sections instead of retrieving the entire conversation at once." That helps you open a long chat. It doesn't cure typing lag in one that's already open. In September 2026, OpenAI support replied on the forum that it had "no timeline to share" for a view that shows only recent messages.
The model side adds time too. Every reply re-reads the whole thread, so each turn sends more input than the last. The fix is the same new chat. Write a handoff while the old one still answers. ChatGPT conversation too long has a paste-ready handoff prompt.
Why is ChatGPT so slow today?
If every chat is slow on every device, it's OpenAI. Check status.openai.com or the free Is ChatGPT down? page Whizi runs, which reads the same feed. OpenAI's help page names a cause it can't fix for you: "ChatGPT users during peak hours may notice a pattern of slowness at specific times of the day." Its Plus help page lists "priority access during high-traffic periods" as a paid benefit, with no numbers.
Don't wait for the word slow. We read the incident data behind status.openai.com on 12 October 2026. It held 110 incidents from 13 July to 9 October 2026. Five had latency in the title, and none of those five touched ChatGPT Conversations. They hit the API, ChatGPT Work and image generation. Of the 35 incidents on Conversations, 31 named errors in the title and none named slowness or latency. OpenAI marked 31 of the 35 degraded performance, not an outage.
The uptime bars miss slowness too. Responses and Chat Completions both show 100% uptime for the 90 days. Yet each one carried two incidents with latency in the title in that window. A green bar there doesn't mean fast.
Timing matters. Of the 35 Conversations incidents, 14 started between 14:00 and 18:00 UTC, which was 10am to 2pm in New York. Those four hours held 40% of the incidents. If ChatGPT drags at the same hour every day, try it earlier. If replies fail outright, ChatGPT network error sorts your connection from theirs.
Why is ChatGPT slow on my computer but fine on my phone?
ChatGPT that's slow in one browser and fine elsewhere points at the browser. Usually it's an extension or stale site data. OpenAI's guide says to try "disabling extensions, especially those that block scripts or ads". A private window turns most of them off, so it's the quickest test. Fast there? Turn your extensions back on one at a time.
For a browser that used to be fine, start with the cache. OpenAI says sudden slowness is "most often" outdated cache data. If the page sticks on Thinking or goes blank, it suggests a hard refresh (Shift and Reload), signing out and back in, and clearing the site data, not just the cache. Its help page still names chat.openai.com, the old address. Clear the data for chatgpt.com too.
The phone app skips the browser entirely, which is why long threads that crawl in Chrome often run fine there. Networks are the other split. OpenAI says a company network or VPN "may have new policies" that get in the way, and mobile data can mean slower replies. Closing other tabs helps an old laptop, since OpenAI says it frees up system resources.
What can you use when ChatGPT is too slow to wait?
Another fast model, with your thread pasted in. What to use when ChatGPT is down lists the other assistants and their status pages.
Whizi, which publishes this page, doesn't make OpenAI's models faster. Two things help when a thread drags. The reply runs on Whizi's server, not in your tab, so closing a frozen tab doesn't kill the answer. And switching models mid thread carries the conversation across (how switching works), so a slow GPT reply can go to Claude Haiku instead. The $15.99 Starter plan includes ChatGPT Luna and Claude Haiku in one picker.
- Switch the thinking level to Instant for questions that need no reasoning
- Open a new chat and send a short prompt to test the thread
- Check the Conversations row on status.openai.com
- Try a private window to rule out extensions
- Clear ChatGPT's site data and sign back in
- Turn off the VPN and try mobile data
Frequently asked questions
Is ChatGPT slower on the free plan?
At busy times, yes, by OpenAI's own description. Its Plus help page lists "priority access during high-traffic periods" and "faster response speeds" as paid benefits, without numbers. The free model itself is quick: Free and Go run GPT-5.6 Luna, which Artificial Analysis timed at 0.7 seconds to its first word with no reasoning. Outside peak hours, a free chat that drags is more likely a long thread or the browser. Is ChatGPT Go worth it compares what that paid plan adds over Free.
Does clearing the cache make ChatGPT faster?
Sometimes. OpenAI says sudden slowness in a browser that used to work is most often outdated cache data, so clearing it is worth one try. It won't speed up a reply that's thinking, a long thread, or an OpenAI incident. If chats look missing after you sign back in, ChatGPT chat history missing lists seven places they hide.
Why is ChatGPT slow all of a sudden?
Check status.openai.com first. A posted incident explains a sudden slowdown on every device. If the page is green, look for a thinking level above Instant, a thread that grew long, or a new browser extension, in that order. Which ChatGPT model should I use matches Instant and the higher levels to the job.
Is ChatGPT slower at certain times of day?
Its incidents bunch up in the afternoon, UTC. Of 35 incidents on ChatGPT Conversations from 13 July to 9 October 2026, 14 started between 14:00 and 18:00 UTC, which is 10am to 2pm in New York. If ChatGPT drags at the same hour every day, try it earlier. ChatGPT statistics shows how many people use it each week.
