Long-document analysis: single-pass Claude now beats RAG below 50 docs
- What happened
- We re-tested document analysis after the 1M context release and split the verdict by corpus size.
- Why it matters
- Below ~50 documents, a retrieval pipeline is now overhead you pay for nothing.
- What to do
- Count your documents: under 50 and one-shot, go single-pass; living corpus at volume, keep RAG.
We re-ran our document-analysis comparison after Claude 4.6 Opus shipped its 1M-token context window.
For corpora under roughly 50 documents, single-pass analysis now beats a retrieval pipeline on both accuracy and setup cost. Retrieval still wins decisively for large or frequently-updated corpora.
The verdict is now split by corpus size instead of a single recommendation — the first time volume, not capability, is the deciding axis.
Retrieval pipeline (RAG)previous pick
Single-pass Claude 4.6 Opusnew pick
The 1M context window removes chunking overhead below ~50 documents. RAG still wins at volume.
Affected tools & models
Never need to catch up again
The weekly delta — only verdict changes and act-now items. No digest filler.