A 1M-token context window is roughly 750 pages in one prompt — and on Claude Opus 5 it is the default, not an opt-in mode. The interesting capability is not the size, though. It is that the model can connect a clause on page 40 to a contradiction on page 600 without you knowing to ask.
This page is written for researchers, operators, and professionals buried in documents. If you want the model overview first, start with the complete Claude Opus 5 guide.
| Best effort level | high |
| Context available | 1M tokens — enough for real working material, not just snippets |
| Cost | $5 per million input tokens, $25 per million output |
Why Claude Opus 5 suits long-document analysis
- 1M tokens by default: no chunking pipeline, no retrieval layer, no lost cross-references.
- Instruction-following and reasoning stay strong across the full window rather than degrading at the far end.
- It reads PDFs including scanned and image-heavy ones, thanks to high-resolution vision.
- It cites where in the document a finding came from when you ask it to.
A workflow that works
- Load everything relevant in one prompt. Splitting the corpus is what creates missed cross-references.
- Ask cross-cutting questions, not page-level ones — contradictions, omissions, and changes between versions are where it beats search.
- Demand citations. "Quote the passage each finding comes from" turns an answer into something checkable.
- Use prompt caching if you will ask many questions of the same corpus — cached reads cost roughly a tenth of the base input price.
- Ask what is missing, not just what is there. Absence is the hardest thing to find with search.
Settings to start from
Set the effort level to high. Long-context work is comprehension-heavy rather than search-heavy. high is the right default; max is worth testing when the stakes justify it.
Leave adaptive thinking on — it is the default on Claude Opus 5, and turning it off introduces failure modes that are not worth the saving. See effort levels explained if you want to tune this properly.
What to watch out for
- Big prompts cost real money. A 500K-token prompt at $5 per million input tokens is $2.50 per question before you have generated a word — cache it if you plan to ask more than once.
- Verify quotes against the source. Ask for exact quotations so verification is trivial.
- Not legal advice. Contract analysis is a drafting and review aid, not a substitute for a lawyer.
The verdict
The clearest single reason to reach for Opus 5 over a cheaper model when your problem is a pile of documents rather than a hard question.
If cost is the constraint, read how to cut Claude Opus 5 costs before downgrading — routing and caching usually beat switching models.
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 many pages fit in Claude Opus 5's context window?
Roughly 750 pages of text, at 1M tokens. On Opus 5 that window is both the default and the maximum — there is no separate long-context mode to enable and no premium tier of pricing for using it.
Is it cheaper to use retrieval instead of the full context window?
Sometimes. A 500K-token prompt costs about $2.50 in input tokens at $5 per million. If you ask many questions of the same corpus, prompt caching brings repeat reads down to roughly a tenth of that, which usually beats building and maintaining a retrieval pipeline for small to mid-size corpora.
Does quality drop at the far end of a long prompt?
Anthropic's guidance is that instruction following, tool calling, and reasoning hold up across the full window on Opus 5. As a practical habit, asking for exact quotations makes any lapse immediately visible and costs you nothing.
