Claude Getting Slower? Why It Happens and How to Fix It

Claude getting slower usually is not a fault in Claude itself—it is the weight of your session. Long conversations, large attachments, extended thinking on hard prompts, and peak-time server load all pile on latency. The quickest fix is to start a fresh chat, trim the context you send, and switch to a lighter model like Haiku or Sonnet.

Model context limits and status-page guidance verified 31 July 2026 against Anthropic's model overview.

If you have used Claude for a while, you know the feeling. The first few replies snap back instantly, then somewhere around message twenty the cursor just... sits there. You are not imagining it, and your internet connection is probably fine. There are a handful of concrete reasons behind Claude getting slower, and the good news is that almost all of them are things you can control. This comes out of genuine time spent with it, so the trade-offs below are ones we have felt firsthand.

This guide walks through every common cause—from a bloated context window to browser tabs quietly eating your memory—and pairs each one with a fix you can apply in seconds. Claude currently spans models such as Opus 5, Sonnet 5, and Haiku 4.5, and we review this guide regularly so the fixes below stay accurate as the lineup changes—most of them take under 30 seconds to try. By the end you will know exactly what to change the moment Claude getting slower starts to interrupt your work.

Why is Claude getting slower?

Claude re-reads the whole transcript on every turn, so a chat that starts at a few thousand tokens and grows past a hundred thousand makes each reply heavier. Peak traffic, extended thinking, big attachments, and a stale browser tab add the rest.

Every time you send a message, Claude re-reads the entire conversation before it writes a single word. That is how it stays coherent: it has no memory between turns except the transcript sitting in front of it. So the transcript itself is the hidden cost. A chat that started at a few thousand tokens can balloon past a hundred thousand by the time you are deep into a task, and each new reply now has to carry all of that weight. There is a hard ceiling, too: per Anthropic's model overview, Opus 5 and Sonnet 5 hold up to roughly 1 million tokens while Haiku 4.5 tops out near 200,000 — and the closer a chat creeps toward that limit, the heavier every turn gets.

That single fact explains most cases of Claude getting slower. But it is not the only factor. Here are the five that matter most:

• A long conversation filling the context window. More history means more for Claude to re-read on every turn. This is the number-one cause of slowdown.
• Peak-time server load. When a lot of people hit Claude at once, requests queue up and responses arrive later than usual.
• Extended thinking on hard prompts. When Claude reasons step by step before answering, it generates extra hidden tokens first, and that takes real wall-clock time.
• Large attachments. A big PDF, spreadsheet, or code file is read into the context and re-processed on every message that follows.
• Browser issues. A tab left open for hours, a flaky network, or a pile of extensions can make the app feel sluggish when the real bottleneck is your own device.

The reason this matters is that the fix depends entirely on the cause. Waiting out a server spike does nothing for a bloated chat, and starting a fresh chat does nothing for a genuine outage. Match the symptom to the right column below.

How to stop Claude getting slower

Work the fixes in order: start a fresh chat, trim what you send, switch extended thinking off for easy prompts, and drop to a lighter tier such as Haiku 4.5. The first two resolve the large majority of cases.

Once you know the cause, the fixes are quick. Work down this list roughly in order—the first two solve the large majority of cases.

1. Start a fresh chat

This is the single most effective move, and it is the one people resist most because they do not want to lose their thread. But a long conversation is exactly what makes each reply heavier. Starting a new chat resets the context back to near zero, and responses snap back to full speed instantly.

If you need to carry context over, do not copy the whole transcript. Ask Claude for a short summary of the important decisions, copy that, and paste it as the first message of a fresh chat. You keep what matters and drop the dead weight. This is closely related to why Claude cuts you off mid-task and why you eventually hit the message limit—all three are symptoms of the same overloaded window.

2. Trim the context you send

If you would rather stay in the current chat, shrink what Claude has to process. Remove large attachments you no longer need. Instead of pasting an entire 2,000-line file, paste only the function you are asking about. Keep questions scoped and specific rather than open-ended. A tight prompt gives Claude less to chew through, which is often the difference between a two-second reply and a twenty-second one.

Understanding how much fits in a session helps here. The context window is finite, and once you are near the top of it, every extra paragraph you add makes a slow reply more likely, not less.

3. Turn off extended thinking when you do not need it

Extended thinking is fantastic for genuinely hard problems—multi-step math, tricky debugging, careful planning. But it is overkill for "reword this email" or "what is the capital of Peru." When it is on, Claude produces a chunk of internal reasoning before it writes your answer, so you wait longer for the first visible word.

If a simple prompt is dragging, switch extended thinking off and watch the delay disappear. You can read exactly how the feature behaves in Anthropic's own extended thinking documentation, and our guide to Claude's extended thinking covers when it is worth the wait and when it is not.

4. Switch to a faster model

Not every task needs the most powerful model. Claude comes in a family of models tuned for different trade-offs, and speed is one of them. Haiku is built to be fast and light; Sonnet balances speed with capability; Opus is the most capable but takes the longest per reply.

If Claude getting slower is hurting a batch of simple tasks, drop to Haiku or Sonnet and you will feel the difference immediately. We route by task all day in our own work — Sonnet as the everyday default, escalating to Opus only for genuinely hard problems, and dropping to Haiku for high-volume, repetitive passes — and the surprise for us was how often the "slow Claude" complaint was really just the wrong model for the job. Reaching for the heaviest model on trivial edits is the most common self-inflicted slowdown we see, and matching the tier to the difficulty fixes it before any context-trimming is needed. Our models explained rundown breaks down each tier, and if you work in the terminal, here is how to change the model in Claude Code with a single command.

Rule out server load and browser problems

If a brand-new short chat is also sluggish, the cause is not your session. Check Anthropic's status page for an active incident first, then suspect the browser: refresh the tab, clear the cache, or disable heavy extensions one at a time.

If a brand-new, short chat is also slow, the problem is probably not on your end at all. Heavy traffic at peak hours adds delay across the board, and occasionally there is a genuine incident. Check Anthropic's official status page before you troubleshoot anything else—if there is an active incident, the only real fix is to wait it out.

When the status page is green but only the web app feels sluggish, suspect your browser. Refresh the tab, clear the cache, disable heavy extensions one at a time, or open Claude in a private window. A stale tab that has been alive for six hours is a surprisingly common reason people think the slowdown is a platform problem when it is really a local one.

Restart the app or your session

The oldest trick still works. Fully close the browser tab or the desktop app, reopen it, and start clean. This clears any stuck front-end state, releases memory, and re-establishes a fresh network connection. If you use Claude Code in the terminal, ending the session and starting a new one does the same job and clears any accumulated context along with it — and if that is where the lag lives, our guide to why Claude Code gets slow walks the terminal-specific causes in the order they pay off.

A quick decision path

Ask five questions in order: is a new short chat slow too, has this thread run long, did an upload trigger it, is extended thinking on for a trivial prompt, and is only the browser affected.

When you next notice Claude getting slower, run through this in order:

1. Is a new short chat also slow? Check the status page—it may be server load.
2. Have you been in this chat a long time? Start a fresh one.
3. Did it slow down after an upload? Remove the big file.
4. Is a simple prompt taking ages? Turn off extended thinking or switch to Haiku.
5. Still slow, and only in the browser? Refresh, clear cache, or restart.

Ninety percent of the time, one of the first two steps fixes it. The whole point is that the slowdown is almost never permanent and almost never a mystery—it is a predictable result of how much you are asking Claude to hold in mind at once. If the slowdown arrived suddenly and everywhere, check Claude AI performance issues to rule out an Anthropic-side incident first.