Claude usually says “try again” when the request hits a limit, fails a safety check, or the model can’t complete the task in its current context. A recent report noted that Claude has also been giving some users repeated “go to sleep” style replies, which Anthropic described as a “character tic” and said it plans to fix in future models.

Why it happens

  • The prompt may be too long or the context window may be close to full, so the model asks for a retry instead of continuing.
  • The request may be blocked by a safety rule, especially if it looks risky, ambiguous, or policy-sensitive.
  • The service may be having a temporary outage or internal error; in that case the system can return a generic retry message.
  • Sometimes the model just falls into a repeated phrasing pattern from training data, which can look odd but is not intentional.

What users are seeing

Reports in May 2026 said many users were posting the same kind of Claude behavior online, including repeated sleep-related responses and “try again later” messages. The public discussion has leaned toward a mix of model quirks, safety behavior, and occasional service issues rather than anything mysterious.

What to do

  1. Shorten the prompt and remove extra context.
  2. Split one big task into smaller steps.
  3. Retry after a minute if the message looks like an error.
  4. Rephrase anything that could trigger a safety filter.
  5. If it keeps happening across many prompts, treat it like a product issue rather than a user mistake.

In plain English: “try again” usually means Claude got confused, hit a limit, or ran into a temporary problem — not that you did something wrong.

TL;DR

Claude’s “try again” message is usually a sign of context limits, safety filtering, or a temporary system issue, and recent reports suggest some of these odd reply patterns are known quirks Anthropic is working on.