Plan limits

Claude context window size by model: 1M, 500K or 200K tokens on paid plans

On paid plans, Opus 5.5, Opus 5, Fable 5.1, Sonnet 5.5 and Sonnet 5 get 1M tokens in chat. Fable 5, Opus 4.6 to 4.8 and Sonnet 4.6 get 500K in chat, and older models 200K. Claude Code gives all of the current models 1M.

Checked against the lab's own pages on October 2, 2026. Limits change often, so the date matters.

The context window is how much text Claude can hold in one conversation, counted in tokens (the chunks of text a model reads, often part of a word). Anthropic keeps the sizes in a help article with three tables, because the same model can get a different window in a chat than it does in Claude Code or Cowork.

Opus 5.5 in chat

1M tokens

Fable 5.1 in chat

1M tokens

Sonnet 5.5 in chat

1M tokens

Fable 5 in chat

500K tokens

Other models in chat

200K tokens

Claude Code

1M on every listed model

Chatting with Claude

On a paid plan (Pro, Max, Team or Enterprise) the newest models get the full 1M tokens in chat: Fable 5.1, Opus 5.5, Opus 5, Sonnet 5.5 and Sonnet 5. The step down is 500K, for Fable 5, Sonnet 4.6 and Opus 4.6 through 4.8. Anything that isn't on the list gets 200K (Anthropic's rough conversion is about 500 pages of text or more).

Fable 5 stands out. It has half of Fable 5.1's window, so if you switch from one to the other partway through a long chat, less of it fits.

You don't get every token for your own text, either. Part of the window is always held back for Claude's reply, so the longest conversation you can have is a little shorter than the table says.

Claude Code and Cowork

Claude Code is simpler. Every model on its list gets 1M tokens, the older Opus 4.6 and Sonnet 4.6 included, though those two come with a catch: you pick the long version with /model (claude-opus-4-6[1m] or claude-sonnet-4-6[1m]) and it needs usage credits turned on. For Opus 4.6 that's only on Pro. For Sonnet 4.6 it's every plan except usage-based Enterprise.

Cowork gives the newest models 1M as well. Opus 4.6, Sonnet 4.6 and Haiku 4.5 drop to 200K there, and Sonnet 5 carries a note of its own: Cowork compacts the conversation automatically once it reaches 500K tokens (half the window).

When a chat fills up

If code execution is on, a paid plan summarizes the early part of a long chat as it nears the limit, and you may see Claude say it's "organizing its thoughts" while that happens. Your full history stays where it was. It isn't free, though. Anthropic says long chats that trigger it use more of your usage limit. With code execution off, you'll run into the length limit and have to start a new chat.

Projects stretch things further on paid plans with RAG (retrieval-augmented generation), which pulls in only the parts of your files a question needs instead of the whole lot. Anthropic's help center says RAG mode can expand a project's capacity by up to 10x. The upload caps that go with that are listed with the other plan limits.