FIX Jun 14, 2026 9 min read

Claude Slow or Not Responding? Fix It Fast

TL;DR Check status.anthropic.com first — most “Claude is slow” reports coincide with server-side degradation. Long conversations and large file attachments are the top user-side cause of slowdowns; start a new…

by Bugi 9 min
TL;DR

  • Check status.anthropic.com first — most “Claude is slow” reports coincide with server-side degradation.
  • Long conversations and large file attachments are the top user-side cause of slowdowns; start a new chat before debugging further.
  • Browser extensions, VPNs, and stale sessions cause the majority of “not responding” freezes in the web app.

Overview

Claude can feel slow or stop responding entirely for reasons that range from Anthropic’s servers being under load to your browser choking on a bloated conversation. The fix depends on where you’re using Claude — the web app at claude.ai, the desktop app, the API, or Claude Code in a terminal.

This guide separates those environments so you can skip straight to the steps that apply. Most issues resolve within the first two fixes. If they don’t, the later steps cover context-window limits, rate throttling, and network-layer problems that are harder to spot.

Quick fixes
  • Check Anthropic’s status page for active incidents.
  • Start a new conversation — long threads hit context limits and slow generation.
  • Hard-refresh the browser tab: Ctrl + Shift + R (or + Shift + R on Mac).
  • Disable browser extensions, especially ad blockers and privacy tools that intercept streaming connections.
  • Switch networks or disable your VPN — corporate proxies frequently throttle streaming responses.

Symptoms

What “slow” and “not responding” actually look like varies by interface:

Web app (claude.ai): The typing indicator spins indefinitely. Responses stream one token per second instead of the usual speed. The “Stop” button appears but nothing renders. The page shows a generic “Something went wrong” banner after a long wait.

Desktop app: The input field accepts text but pressing Enter produces no response. The app window grays out or shows a loading spinner that never resolves.

API / Claude Code: Requests return HTTP 529 (overloaded) or 503. Streaming connections drop mid-response. Claude Code hangs after submitting a prompt, with no output and no error.

Across all interfaces: Partial responses that cut off mid-sentence, noticeably longer time-to-first-token than usual, or repeated “network error” messages on retry.

Why This Happens

Claude slowdowns fall into three buckets, and identifying which one you’re in determines the fix.

Server-side load. Anthropic’s infrastructure handles variable traffic. Peak hours — typically US business hours on weekdays — can degrade response times. Major model launches or viral social-media moments spike demand further. You can’t fix this; you can only confirm it and wait.

Context window saturation. Every message in a conversation accumulates tokens. When a thread approaches the model’s context limit, each response requires processing the full history. Attaching large files, pasting long code blocks, or having 50+ back-and-forth messages in a single chat are the most common user-controllable causes of slowdowns.

Client-side interference. Browser extensions that modify page content (Grammarly, uBlock Origin with aggressive cosmetic filters, privacy badgers), VPNs that add latency or modify TLS connections, and stale authentication sessions all cause symptoms that look like Claude is broken when it’s your local environment.

Note
A conversation that was fast yesterday can be slow today with the same content — context processing cost isn’t linear, and server load fluctuates independently.
01

## Check server status and account access

Open status.anthropic.com. If any component shows “Degraded Performance” or “Partial Outage,” the problem is upstream. Wait it out.

If the status page is green, verify your own access:

– **Log out and log back in.** Stale sessions are a frequent silent cause. On claude.ai, click your profile icon → Sign Out, then sign in again.
– **Check your plan’s usage.** Free-tier users may hit rate limits faster than Pro subscribers during high-traffic periods. If you recently hit a usage cap, Claude may throttle or refuse requests until the window resets.
– **Try a different conversation.** If only one specific thread is slow, the issue is conversation-specific (likely context length), not account-wide.

02

## Try a clean browser or app session

The fastest diagnostic: open an incognito window and log into claude.ai. Send a one-line prompt like “Reply with the word hello.” If you get a fast response, the problem is your main browser profile.

**Browser-specific steps:**

– **Chrome:** Ctrl + Shift + N for incognito. If incognito works, disable extensions one by one in your main profile to find the culprit. Grammarly and similar writing assistants that inject into text fields are the most common offenders.
– **Safari:** Develop menu → Empty Caches, then test. Safari’s Intelligent Tracking Prevention can interfere with SSE connections that Claude uses for streaming responses.
– **Firefox:** Try a fresh profile (`about:profiles` → Create New Profile) rather than just private browsing, since Firefox private mode still loads some extensions.
– **Arc / Brave:** These Chromium forks apply their own content modification layers. Test in stock Chrome to isolate.

**Desktop app:** Quit fully (not just close the window) and relaunch. On macOS, use + Q. On Windows, check the system tray — the app often keeps running in the background.

Tip
If you use Claude daily, pin the incognito test as your first diagnostic step. It rules out the most common cause of “not responding” freezes in under 30 seconds.
03

## Manage context length and file attachments

This is the fix most users overlook, and it’s the one most likely to solve persistent slowness that isn’t server-related.

**Start a new conversation.** Every message in a thread is re-processed with each new prompt. A conversation with 40+ exchanges can push close to the context window limit, making each subsequent response slower.

**Be strategic about file uploads.** Attaching a 200-page PDF means Claude processes that entire document with every follow-up. If you only need pages 15–20, paste the relevant excerpt instead.

**When to start fresh vs. continue:** Continue if you’re mid-task and Claude needs prior exchanges for coherence. Start new if you’re switching topics, responses have slowed, or you’re past roughly 20–30 exchanges.

Warning
Attaching the same large file across multiple messages in one thread doesn’t deduplicate — Claude processes every attachment in the history on each turn.
04

## Run network and platform-specific checks

**VPN and proxy issues.** Corporate VPNs and privacy proxies can interfere with Claude’s streaming connections. Disable your VPN temporarily and test. If Claude works without it, the issue is likely TLS inspection or stream handling.

**DNS resolution.** Switch to a public DNS resolver (8.8.8.8 or 1.1.1.1) if you’re on a corporate or ISP-provided DNS that may be slow or filtering.

**Mobile:** Force-quit the Claude app and relaunch. On iPhone, check that Low Power Mode isn’t throttling network activity. On Android, clear app cache via Settings → Apps → Claude → Storage → Clear Cache.

**Claude Code / API:** Slowness in Claude Code often comes from large project contexts sent with each request. Check your `.claude/` configuration. For API users, enable streaming (`”stream”: true`) for long responses — non-streaming requests have a hard timeout.

~/project

$ curl -s -o /dev/null -w "%{time_total}" https://api.anthropic.com
0.342
$ # Under 1s = network is fine. Over 3s = check VPN/proxy.
05

## Handle API rate limits and HTTP errors

If you’re using Claude through the API (directly or via Claude Code), specific HTTP status codes tell you exactly what’s happening:

– **HTTP 429 (Rate Limited):** You’ve exceeded your plan’s request-per-minute or tokens-per-minute quota. Back off and retry after the `retry-after` header value. Implement exponential backoff in your application code.
– **HTTP 529 (Overloaded):** Anthropic’s servers are at capacity. This isn’t your fault. Retry with backoff. Don’t reduce your rate limit — the server will recover.
– **HTTP 503 (Service Unavailable):** Temporary server issue. Same approach as 529.
– **Timeout with no response:** If you’re not using streaming, long prompts with complex instructions can exceed default client timeouts. Increase your HTTP client’s timeout setting or switch to streaming.

For Anthropic’s rate limits documentation, check the official API reference — limits vary by plan tier and model.

Danger
Do not retry 429 errors in a tight loop without backoff. Aggressive retries can extend your rate-limit window or trigger additional throttling.

If It Still Fails

If none of the above resolves the issue:

  1. Capture the exact behavior. Screenshot any error messages. Note the time (with timezone), which model you selected, and whether the issue is consistent or intermittent.
  2. Test on a different device entirely. If Claude is slow on your laptop and your phone, it’s almost certainly server-side. If it’s only slow on one device, the issue is local.
  3. Check community reports. The Anthropic support site and community forums often have real-time reports from other users during degradation events.
  4. Contact support. For Pro and Team plan users, Anthropic offers direct support channels. Free-tier users can report issues through the help menu in claude.ai.
Takeaway

Most Claude slowdowns are either server-side (check status page, wait) or context-related (start a new chat). Everything else is network or browser interference — and an incognito window test proves it in 30 seconds.

01

Isolate the failing surface

Repeat the same claude slow or not responding check in another browser, private window, or app so you know whether the issue is local or account-wide.

02

Remove local blockers

Temporarily disable extensions, VPN, custom DNS, or network filters, then refresh the session and try again.

03

Escalate only after the basics fail

If the issue follows your account across devices, collect the exact error text and contact the vendor support channel with that evidence.

FAQ

Why is Claude slow only in long conversations?
Every message in a conversation is re-processed each time you send a new prompt. As the thread grows, the total token count approaches the model’s context window limit, and processing that full history takes progressively longer. Starting a new conversation resets this overhead immediately.
Does Claude have a rate limit?
Yes. Both the free tier and paid plans have rate limits based on requests per minute and tokens per minute, though paid plans have significantly higher thresholds. When you hit a limit, Claude returns an HTTP 429 error or shows a usage-cap notice in the web app. Limits reset automatically after a short window.
Why does Claude stop mid-response?
Mid-response cutoffs usually happen because the output hits the model’s maximum output token limit, or a network interruption breaks the streaming connection. Try asking Claude to continue from where it stopped, or break your request into smaller parts so each response stays within the output limit.
Is Claude slower on the free plan?
Free-plan users share a more limited capacity pool and may experience throttling or queuing during peak traffic periods. Pro and Team subscribers get higher rate limits and priority access, which generally results in faster and more consistent response times.
How do I check if Claude is down?
Visit status.anthropic.com for real-time information on API availability, web-app health, and any active incidents. You can subscribe to email or RSS updates on that page to get notified automatically when outages occur.
Why is Claude Code slower than the web app?
Claude Code sends your project context — file contents, directory structure, and conversation history — with each request, which can add significant token overhead. Check your .claude/ configuration to make sure you’re not accidentally including large directories, and keep individual prompts focused to reduce the per-request context size.