Why is ChatGPT so slow? The cause is usually one of six things: an OpenAI service incident, a reasoning-heavy model or mode, an oversized conversation, a file or tool workflow, a local browser or network problem, or an account or regional restriction. Check those layers in that order. The fastest useful test is a short prompt in a new chat: if that works, the old conversation or task is the likely bottleneck; if it does not, move on to the service, browser, network, and account checks below.
Current guidance checked July 21, 2026: OpenAI’s official developer documentation says lower reasoning effort favors speed, while higher effort spends more work on difficult tasks. It also treats output length, repeated requests, large inputs, and tool use as separate sources of latency. Those principles explain why a complex task can be slow even when ChatGPT is not experiencing an outage, but API settings do not prove which model or configuration your ChatGPT account is using.
Is ChatGPT Down or Just Slow? Start Here
| What you see | Most useful first test | What the result suggests |
|---|---|---|
| Plain prompts are slow on multiple devices or networks | Open the OpenAI Status page | A listed ChatGPT incident points to an OpenAI-side problem; local browser fixes are unlikely to resolve it |
| Only one long conversation is slow | Run a short prompt in a new chat | The original thread is probably carrying too much context or browser-rendering load |
| The interface stays in a thinking phase on a hard task | Try a simpler prompt or a faster option that your interface actually offers | Deliberate reasoning may be the cause rather than an outage |
| Text chat is normal, but files, research, images, or Canvas are slow | Repeat the task without the file or tool | Upload, parsing, retrieval, tool execution, or rendering is the extra step |
| Typing, scrolling, or buttons lag before a request is sent | Try a private window with extensions disabled | The browser, extension stack, or local device is the likely cause |
| You see a limit, access, workspace, or region message | Read the exact notice and compare with the same account on another supported device or network | This is an entitlement or policy issue, not ordinary response latency |
If ChatGPT fails rather than merely responds slowly, use the separate guide to diagnose why ChatGPT is not working. A blank page, login loop, explicit error, missing feature, and slow answer are different symptoms and should not be treated as one problem.
How to Fix Slow ChatGPT, From Fastest Check to Support
1. Check OpenAI’s live service status
Start with status.openai.com when the slowdown began suddenly, affects ordinary text prompts, or appears on more than one device. Look for a current ChatGPT incident, degraded performance, or a component-specific notice. If OpenAI lists an active incident, preserve your work and wait for the status update rather than repeatedly clearing data or changing account settings.
2. Test a short prompt in a new chat
Open a fresh conversation and send a small, plain-text request such as “Summarize this sentence in five words.” Do not attach a file, enable research, or request a long answer. If the fresh chat is responsive, the old thread or complex task is the likely cause. Copy only a concise summary of the necessary context into the new chat; do not move the entire history unless it is essential.
3. Use less reasoning for a simple task
If your ChatGPT interface offers a faster model, a lower-thinking option, or a way to reduce reasoning depth, use it for rewriting, extraction, short summaries, and routine questions. Keep deeper reasoning for debugging, planning, research, or other tasks where the quality trade-off matters. OpenAI’s current reasoning documentation states the underlying principle clearly: lower effort favors speed and lower token use, while higher effort allows more complete reasoning.
Do not assume that a model name or setting described in API documentation is available in your ChatGPT picker. Product access can depend on the plan, workspace, rollout, and interface you are actually using.
4. Remove files, research, images, and other tools
Retry the same goal as a plain-text chat. File uploads must be transferred and parsed; research workflows may search and synthesize multiple sources; image tasks need separate generation or analysis; and Canvas must render an interactive document. OpenAI’s deep research documentation, for example, describes a workflow that can combine web search, file search, remote tools, and analysis. That is not equivalent to a single text response.
If only Canvas is sluggish, save the important text, open a smaller document, and use the ChatGPT Canvas guide to separate document-rendering issues from model response delays.
5. Isolate the browser and extensions
- Save unsent work, then reload the page.
- Open ChatGPT in a private or incognito window.
- Temporarily disable content blockers, script injectors, writing assistants, privacy extensions, and antivirus browser add-ons one at a time.
- Try a current version of another supported browser.
- If the page itself is heavy, close unused tabs and restart the browser.
If private browsing fixes the problem, an extension, cookie, cached asset, or browser profile is the likely cause. Clear site data only after saving work and confirming that you can sign in again. On a managed work or school device, do not disable required security controls without administrator approval.
6. Compare networks without bypassing access rules
Try a second trusted connection, such as home broadband versus a mobile hotspot. If one network works and the other does not, investigate Wi-Fi quality, DNS filtering, proxy inspection, firewall rules, or VPN routing. Disconnect a personal VPN for the test if doing so is safe. Do not use a VPN to bypass regional availability, workplace policy, or account restrictions; contact the network administrator or OpenAI Support instead.
7. Separate usage limits from slow responses
A visible usage-limit, reset-time, unavailable-model, workspace-policy, or unsupported-region notice is not a speed measurement. Follow the exact notice shown in your account. Compare the same short task in a fresh chat using an option that is visibly available to you. Avoid relying on old screenshots or fixed quota figures because limits and entitlements can vary by product, plan, workspace, and time.
8. Collect evidence and contact OpenAI Support
If the problem survives all earlier tests, open the OpenAI Help Center and use its support channel. Include the time and timezone, the exact symptom or error, the affected feature, browser and operating-system versions, network type, whether a VPN or extension was involved, and the tests that changed or did not change the result. For an account-specific issue, include the account email only through the official support flow.
Do not post passwords, API keys, private files, or confidential conversation contents in a public forum. Browser archive files and console logs can contain sensitive data; create or share them only when official support requests them and after reviewing what they contain.
Why ChatGPT Can Be Slow When the Service Is Healthy
Reasoning depth changes the speed-quality trade-off
A reasoning phase is model work, not necessarily a frozen page. Complex coding, multi-step analysis, research, and planning can justify more work before the answer. A routine rewrite or extraction usually does not. The practical test is not an arbitrary wait-time threshold; it is whether a simpler task, fresh chat, or lighter available setting responds normally.
Long conversations create two different bottlenecks
First, the browser must display and manage a larger page, which can make typing and scrolling lag. Second, the request may carry more relevant context for the model to process. Starting a fresh chat tests both problems at once. Preserve decisions and essential facts in a compact handoff summary, then leave obsolete turns behind.
Files and tools add processing stages
A tool-enabled request can include upload, indexing, retrieval, web access, code execution, image work, or document rendering before the final answer appears. If plain chat is responsive but a tool workflow is not, troubleshoot that feature rather than treating all of ChatGPT as slow.
The browser can be slow even before the model receives anything
If keystrokes, menus, or scrolling lag before you submit a prompt, the model is not the first suspect. Browser memory pressure, extensions, cached code, device load, or a very long rendered thread better matches that symptom.
Network and account problems can look similar
A network problem often changes when you switch connections. An account or workspace problem usually follows the signed-in account and may display a limit or policy notice. Testing one variable at a time prevents unnecessary changes and gives support a useful record.
For Developers: Diagnose API Latency Separately
Browser troubleshooting does not diagnose an API request, so developers should measure API latency separately. As of July 21, 2026, OpenAI’s official reasoning guide identifies gpt-5.6 as the starting point for most reasoning API workloads and documents model-dependent effort levels. It also describes a pro reasoning mode for difficult work that can tolerate higher latency and token use. These are API facts; they should not be used to infer a ChatGPT plan’s picker or limits.
- Log time to first byte, time to first streamed token, total duration, input size, output size, model, reasoning settings, and tool calls separately.
- Treat an explicit rate-limit response as a rate-limit problem. OpenAI’s rate-limit guide documents request, token, project, organization, model, and usage constraints plus response headers for remaining and reset values.
- Use streaming to improve perceived responsiveness, while recognizing that it does not remove the model’s total work.
- Request shorter outputs when a long answer is unnecessary, reduce avoidable sequential calls, and remove genuinely irrelevant context.
- Use OpenAI’s latency optimization guide as the implementation checklist instead of relying on a universal speed benchmark.
When a Separate Multi-Model Workspace Helps
Disclosure: GlobalGPT publishes this Hub article and operates the service described in this section. GlobalGPT is a separate third-party multi-model platform, not an OpenAI product. Its interfaces, model access, privacy terms, limits, and outputs can differ from ChatGPT.
A separate provider can be useful when OpenAI has a confirmed incident or when you want to compare how another model handles a non-sensitive task. It will not repair an OpenAI login, billing, workspace-policy, or account-limit problem. Before moving work, remove secrets and personal data, confirm that the destination is allowed by your employer or client, and review the platform’s current terms. If you need a broader comparison, see the guide to all-in-one AI tools and platforms.
Frequently Asked Questions
Why is ChatGPT suddenly so slow today?
A sudden slowdown across ordinary prompts, devices, or networks can indicate an OpenAI incident. Check the live status page first. If no incident is listed, test a short prompt in a new chat, then isolate the browser, network, files, tools, and account in that order.
Does a long ChatGPT conversation make replies slower?
It can. A long thread can make the browser render more content and may give the model more relevant context to process. Summarize the necessary state and continue in a fresh chat. If the new chat is faster, the old thread was a meaningful part of the slowdown.
Why does ChatGPT stay in a thinking phase?
The selected task or available mode may be using more reasoning before answering. Try the same goal with a simpler prompt or a faster option shown in your interface. Do not use a fixed number of seconds as proof of failure; compare behavior across a plain prompt, fresh chat, and live status notice.
Why are file uploads or Deep Research slower than normal chat?
They involve additional stages. A file may need upload, parsing, indexing, and retrieval, while a research workflow can search, analyze, and synthesize sources. Retry without the file or tool. If plain text works, troubleshoot the specific workflow rather than the whole service.
Can browser extensions make ChatGPT slow?
Yes. Extensions that block scripts, rewrite text, inject controls, filter traffic, or inspect pages can interfere with the interface. Test a private window and disable extensions one at a time. Do not disable organization-required security controls without permission.
Can a VPN or corporate network cause ChatGPT lag?
Yes. Distant VPN routes, proxy inspection, DNS filtering, and firewall rules can delay or interrupt connections. Compare a second trusted network. Do not use a VPN to bypass account, workplace, or regional restrictions; ask the administrator or OpenAI Support for help.
Is a usage-limit message the same as ChatGPT being slow?
No. A usage-limit or unavailable-feature notice is an account, plan, workspace, or product-access condition. Follow the exact reset or access information displayed in your account. Old quota tables and screenshots should not be treated as current evidence.
When should I contact OpenAI Support about slow ChatGPT?
Contact support when no relevant incident is listed and the issue persists in a new chat, private window, second browser, and second trusted network. Through the official Help Center, send timestamps, versions, exact symptoms, and completed tests—never passwords, keys, or confidential data.
Bottom Line
Do not diagnose ChatGPT speed from the wait alone. First rule out an OpenAI incident, then compare a short prompt in a new chat, reduce unnecessary reasoning and tools, isolate the browser and network, read any account notice literally, and escalate with evidence. That sequence finds the responsible layer faster and avoids destructive or unsafe troubleshooting.
