Error messages

This chat is too long to continue. Start a new chat, or turn off some connectors to free up space.

This chat is too long to continue. Start a new chat, or turn off some connectors to free up space.

This is the Claude app's wall at the end of a chat. On claude.ai and in the desktop app it reads:

This chat is too long to continue. Start a new chat, or turn off some connectors to free up space.

Until recently the same box said This conversation is too long to continue. Start a new chat, or remove some tools to free up space., and that older line is still what most people paste into search. We fetched the claude.ai web code on September 25, 2026 and found the current wording there, shown whenever the model answers with an invalid_request_error containing prompt is too long. That's the same refusal Claude Code gets, but the app handles it differently (more on that below).

What's actually too long

Every turn sends the whole chat to the model again, along with your project instructions, any files you attached and the definitions of every tool Claude can call. The model has a context window (the most text it can read at once), and when that bundle doesn't fit, the chat stops. Anthropic's help article on length limits says the newest models on paid plans take up to 1M tokens and others 500K or 200K, with part of that always kept back for Claude's reply.

The chat history isn't always the main cost. The same article says plainly that "Tools and connectors are token-intensive", and the reports on GitHub bear that out.

When a new chat hits it on the first message

This is the case that confuses people most. In a June report, a claude.ai user with about 15 connectors counted roughly 500 tool definitions, some 90,000 to 100,000 tokens, loaded before anything was typed. One connector (PostHog) brought about 270 tools on its own. They had turned on "Load tools when needed", which should have kept those definitions out until needed, and it didn't for them. A later comment says it only happened with Opus 4.7 and 4.8 in chat, not in Cowork or Claude Code.

A May report from the desktop app describes every new chat inside one project failing at about 16,000 of 200,000 tokens, even after cutting project instructions down to 80 lines. Nobody from Anthropic replied to either one, as far as we can see (the second was closed for inactivity).

If a fresh chat fails, go to your connectors and switch off the ones this chat doesn't need, then try again. That's what the error's own wording points at.

Keeping long chats alive

For a chat that simply ran long, the help article describes automatic context management: when a chat nears the limit, Claude summarizes earlier messages and carries on, and the full history stays available. It only works with code execution turned on. The article also says very large first messages can still hit the limit. When summarizing can't free enough room, the app shows This chat can’t be compacted any further. Start a new chat to continue.

Anthropic's other suggestions: put big reference material in a project (projects retrieve the relevant parts instead of loading everything), keep project instructions short, remove project files you no longer use, turn extended thinking off when you don't need it and pick a lower effort level for routine work.

It hasn't always rescued people. An April report on desktop 1.2581.0 described a hard stop with no warning and no option to compact, so a long agent session was simply over. If you run long agent sessions, Claude Code's manual compaction is the safer home for them.

Claude Code is a different page

In Claude Code the same refusal shows as Prompt is too long and you can run /compact yourself. That's covered in “Prompt is too long” in Claude Code. If the app instead complains about one message being too big before it even sends, see “Your message will exceed the length limit”.