Reducto r-1: Single-Pass Document Parsing at 1 Cent Per Page

Reducto last week announced r-1, a new document parsing model that replaces the company's multi-stage agentic OCR pipeline with a single full-page pass. According to Reducto, r-1 cuts error rates by 20% compared to its own legacy models and reduces per-page pricing from 3-6 cents down to a flat 1 cent. It is available now in preview through the V3 Parse API, with no open weights and no self-hosted option. The model handles text, tables, figures, layout, reading order, and grounding in one pass.

What happened

Detail	Fact
Model name	r-1
Error rate improvement	20% reduction vs. Reducto’s own legacy agentic pipelines
New price	1 cent per page, flat, all features included
Legacy price	3 to 6 cents per page depending on workload
Speed vs. legacy	Up to 6x cheaper; also faster
Access	Preview via V3 Parse API, flag: settings.model: “r-1”
Open weights	No
Migration credits	Up to $5,000 for side-by-side comparisons from another parser

Reducto’s previous Parse product ran OCR, layout detection, and post-processing as separate stages, then optionally added agentic vision-language passes on top. Each additional model call added latency and cost. r-1 folds all of that into a single pass over the full page image.

Every content block that r-1 returns includes page-relative bounding boxes, so downstream code can tie any extracted value back to its exact position on the page. That matters when auditors or agents need to verify a figure against the source document.

What r-1 parses in one pass

According to Reducto’s documentation, the single pass covers:

• Digital text, scans, and handwriting
• Table structure with surrounding context, including merged cells and nested headers
• Columns, headers, footers, sidebars, and reading order resolved together
• Figure detection with a short generated description
• Formatting that carries meaning: headings, lists, bold, underlines, and strikethroughs
• Page-relative bounding boxes for all blocks

Reducto specifically calls out dense tables, unusual layouts, low-quality scans, watermarked content, and template-free documents as the edge cases r-1 targets. A missed strikethrough can flip the meaning of a contract clause. A misread table cell hands the wrong number to an agent. These are the failure modes that cost real money in regulated industries.

How does r-1 compare to incumbent tools?

Reducto says r-1 outperformed Amazon Textract and Azure Document Intelligence, as well as large general-purpose LLMs, on complex documents in its internal evaluations. Those are vendor-run tests with no public dataset or eval harness released alongside the announcement, so treat the comparative claims as directional rather than verified.

The 20% error reduction is also measured against Reducto’s own prior pipeline, not against any independent third-party baseline. That said, the pricing shift from 3-6 cents to 1 cent per page is a concrete, checkable fact.

Migration and what stays on the old stack

r-1 requires the V3 API. A Parse request that omits the model setting still runs on legacy Parse, so existing integrations do not break silently. New pipelines created in Reducto’s Studio default to r-1.

The basic API call looks like this:

result = client.parse.run(input=upload.file_id, settings={“model”: “r-1”})

Agentic processing is not removed. Workflows that need custom prompts or advanced chart extraction still route those pages through the agentic pipeline, which Reducto layers onto the r-1 result. That adds latency, so teams should only use it where the base model falls short. Reducto recommends checking the r-1 configuration compatibility page before migrating, since some legacy settings are ignored or unsupported under the new model.

Two more products are planned: r-1 mini for speed and cost-sensitive workloads, and automatic per-page routing that picks the right model for each page in a document.

Reducto’s platform supports multi-tenant cloud, customer VPC, on-premises, and air-gapped installs. SOC 2 Type II attestation and HIPAA processing are available on higher tiers, per the company’s security policies.

Why it matters

Teams processing financial statements, insurance claims, or contracts frequently route files through several providers and bolt on post-processing scripts to reach acceptable accuracy. That orchestration adds cost, latency, and maintenance burden. r-1’s pitch is that one model at one price handles the whole job, which simplifies the pipeline rather than just improving one metric.

For businesses already building AI integration workflows on top of document data, a cheaper and more accurate parser at the ingestion layer has a compounding effect. Errors that enter a pipeline at the parsing stage tend to surface late, when they are expensive to fix.

The flat rate card also matters for cost forecasting. When per-page price depends on which features are enabled, budgeting for document volume becomes genuinely hard. One cent per page regardless of complexity is easier to model.

Our take

The architecture change is credible. Collapsing a multi-step pipeline into a single pass is a well-established way to reduce compounding errors, and the pricing move is aggressive enough to force a response from incumbents.

That said, the 20% improvement is self-reported against Reducto’s own prior system, and the hyperscaler comparison has no public benchmark attached. If you are evaluating this for a regulated pipeline, the $5,000 migration credit offer for a side-by-side comparison is the right place to start. Run your own documents through both and measure on output that matters to your workflow, not on Reducto’s internal eval.

For teams already deep in workflow automation that ingests PDFs, contracts, or scanned forms, r-1 is worth a test in preview. The bounding box grounding in particular is useful if your downstream agents need to cite or verify a source position. Watch for the r-1 mini release if cost is the primary constraint and document complexity is lower. You can keep an eye on related developments in our AI news coverage.

What to do about it

1. Check whether you are on the V3 Parse API. If not, upgrade before testing r-1.
2. Review the r-1 configuration compatibility page to identify any legacy settings your pipeline uses that may not carry over.
3. Run a sample of your actual documents through r-1 in preview and compare error rates against your current parser.
4. If you are migrating from another provider, request the up-to-$5,000 credit for a side-by-side comparison before committing.
5. Defer agentic layers to only the pages that need custom prompts or advanced chart extraction, to keep latency low.

A flat 1 cent per page is easy to budget against. Test on your worst documents first, not your average ones.

Source: Marktechpost