claude/for
Search
Subscribe
Work & Tools

Claude taking longer than usual? What the retry means

"Taking longer than usual. Trying again shortly" means Claude is retrying on its own, usually under heavy demand. Wait, then check status.claude.com.

"Taking longer than usual. Trying again shortly (attempt 2 of 10)." That banner means your message did not get an answer on the first try, and Claude is retrying by itself. It is usually a busy-servers problem, not your account or your network. The fastest fix is to let a few attempts run, then check status.claude.com if the counter keeps climbing.

This page explains what the counter means, how it differs from an outage, and what is actually documented about peak hours. claude-for is an independent guide, not affiliated with Anthropic. Sources were checked October 22, 2026.

TL;DR: The banner is an automatic retry, not an error you caused. Wait through a few attempts. If it reaches the end of the count, start a new chat or switch models. If status.claude.com shows an incident, wait it out.

What the message looks like

The wording users post is "Taking longer than usual. Trying again shortly (attempt N of 10)", where N counts up from 1. You can see it verbatim in user posts on X in February 2026 and on Threads at attempt 4 of 10 and attempt 9 of 10.

It appears inside the chat, under your message, while the reply is pending. Some third-party guides quote shorter or reworded variants, such as "(Attempt 3)" with no total, or "This is taking longer than usual. We'll retry shortly." We could not find a screenshot or user post confirming those, so treat the "attempt N of 10" form as the one to search for.

Why you see it

Anthropic has no help article for this exact banner. The closest official explanation is for a sibling message, "Due to unexpected capacity constraints, Claude is unable to respond to your message." Anthropic's error messages article says capacity issues are "not outages": the system "is functioning normally but managing high demand across all users."

The "of 10" counter lines up with how Claude Code behaves. Anthropic's Claude Code error reference says it "retries transient failures up to 10 times with exponential backoff before showing an error." In the terminal that shows as Retrying in Ns · attempt x/y. Anthropic does not say the chat app uses the same logic, so the match is our reading, not a documented fact.

Other causes users and third-party guides point to: a very long conversation, a huge pasted document, or a flaky connection that drops the reply mid-stream. Those are plausible and worth ruling out, but they are user reports, not Anthropic's explanation.

Retry banner vs outage

The two look similar from your chair. They are not the same thing, and the status page only tells you about one of them.

"Taking longer than usual"Outage or incident
What is happeningYour request is being retried, usually under loadSomething is broken on Anthropic's side
Shows on status.claude.comOften notYes, as a named incident
Typical outcomeThe reply arrives after a few attemptsErrors until Anthropic posts a fix
What to doWait, then retry or switch modelsWait, and follow the incident updates

That middle row matters. Anthropic states that "Capacity issues will not appear on our status page because they represent normal load management rather than technical problems." So a green status page does not prove the slowness is on your side.

How to check status.claude.com

Open status.claude.com. The top line reads "All Systems Operational" when nothing is posted. Below it, each product has its own row: claude.ai, Claude Console, Claude API, Claude Code and Claude Cowork. Check the row for the product you are using, not just the headline.

Scroll down for recent incidents. Titles name the symptom, for example "Degraded performance" or "Elevated errors", and often name a specific model. If the affected model is the one you are using, that is your answer. The Subscribe to Updates button sends incidents to email, SMS, Slack, Microsoft Teams, a webhook or RSS.

The Claude moves worth stealing, by email

Workflows tested on real jobs, sent only when they are worth it. Free.

How to fix it

now

Let a few attempts run

The retry is automatic, and user posts show it often clears before the count runs out. Don't resend the same message while the counter is running, or you may get two replies.

next

Check status.claude.com

Look at the claude.ai row and recent incidents. If your model or product is named, wait for the fix.

next

Switch models

Anthropic's Claude Code docs say capacity is tracked per model. Picking another model in the model menu can get you an answer while one model is busy.

next

Start a fresh chat

If the thread is very long, open a new chat and paste only what Claude needs. This also saves usage.

next

Rule out your connection

If everything else fails, try a different network or the browser instead of the app. Our guide to the Couldn't connect to Claude error covers the network checks.

If the counter reaches 10 and the reply still fails, resend the message or start a new chat. If nothing works for an hour and the status page stays green, our broader guide to Claude not working walks through the remaining checks.

How to make it rarer

  • Work outside the busiest hours when you can. Anthropic says capacity problems "typically resolve as demand patterns shift throughout the day."
  • Keep conversations short. Anthropic's usage limits article says longer conversations consume more of your usage, and it recommends starting a new conversation in long chats.
  • Split giant pastes. Send a long document in sections, so each request is smaller.
  • Have a fallback model. If one model is struggling, a lighter one often answers.

What Anthropic says about peak hours

Anthropic does not publish a "slow hours" schedule. What it has documented is about usage limits, not speed. In March 2026, an Anthropic post on Reddit, confirmed by PCWorld, said five-hour session limits were being adjusted during weekday peak hours, 5 a.m. to 11 a.m. Pacific, to "manage growing demand for Claude."

On May 6, 2026, Anthropic announced it was "removing the peak hours limit reduction on Claude Code for Pro and Max accounts," in its higher limits post. That post does not mention chat on claude.ai. If you hit your allowance rather than a slow reply, see our guide to Claude's usage limits.

Anthropic has not published response-time figures by hour, so any claim that a given hour is "slow" is a user observation, not a documented fact.

The honest limits

Anthropic has no help article for "Taking longer than usual. Trying again shortly." Everything about the attempt counter comes from user posts, and the link to Claude Code's 10-retry logic is an inference. We have not reproduced the banner ourselves.

We also cannot tell you how long a given slowdown will last. Capacity problems do not get status-page entries, so there is no official record of past slow periods. Third-party guides that give a fixed waiting time are guessing.

Questions people ask

Why is Claude so slow today?

Most often, demand is high and Claude is retrying requests. Check status.claude.com for a named incident. If none is posted, Anthropic says capacity issues are normal load management and do not appear there. Waiting a few minutes or switching models usually helps.

What does "attempt 2 of 10" mean on Claude?

It is the retry counter. Your first attempt did not return a reply in time, and Claude is trying again by itself, up to 10 times. You don't need to do anything during the count. If all 10 fail, resend the message or start a new chat.

Is Claude down, or just slow?

Check status.claude.com. A named incident means something is broken, and you should wait for the fix. A green page with the retry banner usually means heavy load, which Anthropic says it does not post. Our Couldn't connect to Claude guide covers the case where nothing loads at all.

What are Claude's peak hours?

Anthropic named weekdays, 5 a.m. to 11 a.m. Pacific, when it adjusted session limits in March 2026. That window was about usage limits. Anthropic has published no peak schedule for response speed, and it removed the peak reduction for Claude Code on Pro and Max in May 2026.

When Anthropic changes limits or capacity in a way that affects your day, the newsletter sends one short note.

The Claude moves worth stealing, by email

Workflows tested on real jobs, sent only when they are worth it. Free.