r-1 mini

r-1 mini is a lightweight document parsing model launched on 2026-11-04 by Reducto. It is a smaller sibling of Reducto's unified parser r-1 (full agentic parsing at $10 per 1,000 pages).

r-1 mini brings the price down to $1 per 1,000 pages, and Reducto says it is more accurate than every other cost-effective parser and even premium alternatives costing up to 25× more. Reducto also introduced the ParseArena benchmark and PAIR (Pairwise AI Ranking), an open-source pairwise AI judge framework.

What problem it solves

RAG, research agents and enterprise search depend on turning PDFs, scans and tables into structured data. High-quality parsing is expensive, so many lower-priority documents are skipped.

Highlights

• Price: $1 per 1,000 pages, one tenth of r-1.
• Quality: in Reducto's evals, the next higher-scoring dedicated parser costs 10× more, and the next-cheapest full-feature alternative costs 3.75× more yet scores 16 points lower.
• Evaluation: ParseArena uses diverse enterprise documents judged by PAIR, which compares two parser outputs against the original page instead of rigid scoring rules.
• Keeps some agentic capabilities.

Limitations

• Benchmarks are vendor-built and should be independently checked.
• Closed commercial API, not open weights.
• The hardest layouts may still need full r-1.

Who it's for

• Teams parsing large document archives for RAG and knowledge bases;
• Developers building high-page-volume apps such as web crawling and research agents.

Getting started

1. Read the r-1 mini launch post and sign up for a Reducto API key.
2. Run r-1 mini on bulk lower-priority documents and keep r-1 for critical ones.
3. Use the PAIR/ParseArena approach to run pairwise evals on your own samples and confirm the quality/cost tradeoff.