Measured by us

RAG over documents — Claude Opus 5.5 on Direct API (global)

Results

Answer accuracy	91.2% higher is better
Citation faithfulness	95% higher is better
Hallucination rate	2.1% lower is better

Run details

Model version: claude-opus-5-5
Endpoint: api.anthropic.com /v1/messages, effort=high
Platform / region: Direct API / global
Dataset: docs-qa-500 (500 questions over 120 technical PDFs)
Harness: GitHub @ a1b2c3d
Run cost: $18.40
Date: 23 Sept 2026

What was NOT tested

Non-English documents
Scanned PDFs needing OCR
Retrieval quality (fixed retriever used for all models)

Re-run it on your data

git clone https://github.com/mallimatla/uptodate.git && cd uptodate && git checkout a1b2c3d
npm install && npm run proof -- --suite rag --model claude-opus-5-5 --platform direct --region global --dataset ./your-data

Sample raw output

The retry budget defaults to 3 attempts with exponential backoff [doc 14, p. 7].