Claude Opus 5 has a 1M-token context window, and on this model that figure is both the default and the maximum. There is no long-context mode to enable, no beta header, and no premium pricing tier for using the whole thing.
What 1M tokens actually holds
| Content | Roughly |
|---|---|
| Plain English text | ~750,000 words, or ~750 pages |
| Source code | A large repository with tests and docs |
| PDFs | A few hundred pages, depending on layout and images |
| Meeting transcripts | Dozens of hours of conversation |
Output is capped separately at 128K tokens per response — enough to emit an entire refactored module or a full financial model in one go.
Size is not the interesting part
Plenty of models advertise big windows. What matters is whether quality holds across them — whether the model can connect a clause on page 40 to a contradiction on page 600, or notice that a function on line 8,000 breaks an assumption made on line 200.
Anthropic's guidance for Opus 5 is that instruction following, tool calling, and reasoning stay strong across the full window. The practical habit that makes this verifiable at no cost: ask for exact quotations. If a citation is wrong you will see it immediately.
The cost of a large prompt
A full window costs real money. At $5 per million input tokens, a 1M-token prompt is $5 before the model writes a word. Three practical consequences:
- Cache aggressively. If you ask multiple questions of the same corpus, cached reads cost roughly a tenth of the base rate. This is usually the difference between a workable product and an expensive one.
- Don't fill it out of habit. Sending more context than the task needs is a direct cost with no quality guarantee.
- Consider retrieval at scale. For very large or frequently-changing corpora, retrieval still wins. For small and mid-size document sets, the full window is simpler and often cheaper than the pipeline you would otherwise build.
Context editing and compaction
For conversations that would exceed even 1M tokens, two server-side features help. Compaction summarises earlier context automatically as the conversation approaches a threshold. Context editing clears stale tool results and thinking blocks rather than summarising them. Both are beta features on the API, and both matter mainly for long-running agents rather than ordinary chat.
More on Claude Opus 5
Start with the complete Claude Opus 5 guide for the overview, or go deeper:
Ready to go deeper?
Read the full Claude Opus 5 guideFrequently Asked Questions
How big is Claude Opus 5's context window?
1M tokens, which is both the default and the maximum — there is no separate long-context mode to enable. That is roughly 750,000 words or 750 pages of text. Maximum output is 128K tokens per response.
Does the whole 1M-token window cost extra?
There is no long-context premium on the per-token rate, but you pay for what you send: a 1M-token prompt costs $5 in input tokens at standard pricing. Prompt caching brings repeat reads down to roughly a tenth of that.
Should I use the context window or a retrieval system?
For small to mid-size corpora that you query repeatedly, the full window plus prompt caching is simpler and often cheaper than building and maintaining retrieval. For very large or rapidly-changing corpora, retrieval still wins. Many production systems combine both.
