Reducto Review

Three named customers — Harvey, Scale AI, Vanta — plus 99% recall claims on micro1's benchmark. That's not vaporware. Deep Extract and the citation/bounding-box layer solve a real trust problem in RAG pipelines: knowing where a number came from.

No public funding data, no headcount, no time-in-market disclosed. Category is crowded — Textract, Azure Document Intelligence, and half a dozen IDP startups fight here. Reducto's pitch is replacing 4-5 vendors with one API, which advances your stack rather than just cutting an invoice.

15,000 free credits, then $0.015/credit — cheap enough to pilot without a procurement fight. SOC 2, HIPAA, air-gapped deployment options check enterprise boxes early. The board won't flinch at the cost. They should ask about the company's balance sheet before you standardize.

Five capabilities under one API — Parse, Split, Extract, Classify, Edit — is the right shape for how document-heavy teams actually operate. Most orgs I'd oversee are running OCR from one vendor, extraction from another, classification homegrown. Reducto's pitch to replace 4-5 stitched vendors is a real operational simplification, not just marketing.

The compliance stack — SOC 2, HIPAA BAAs on Growth tier, zero data retention by default, air-gapped deployment on Enterprise — tells me this was built for regulated verticals from day one, not retrofitted. Customers named include Harvey and Vanta, both compliance-sensitive buyers. That's the right signal for finance, healthcare, legal deployments.

15,000 free credits, then $0.015/credit on Standard — fine for piloting, opaque for budgeting at scale since Growth and Enterprise pricing isn't published. Competing against traditional OCR/IDP vendors and parsing APIs, Reducto's 12+ model orchestration removes the burden of tracking frontier model releases yourself. Three-year risk: you're dependent on their routing logic and model relationships, not just their API surface.

15,000 free credits, then $0.015 each. Standard tier math is visible, no sales call needed. That's rare in document AI — category norm is quote-only from page one.

But Growth tier, the one with HIPAA BAAs and priority support, has no published price. Enterprise is fully custom. A team of 50 processing real volume lands in Growth territory fast, and that's exactly where the pricing page goes dark. Compare to Textract or Azure Document Intelligence, both consumption-priced and published per-page — Reducto's credit system adds a conversion layer procurement has to model blind.

No stated contract term or auto-renewal language in the evidence. No overage rate beyond Standard. Deep Extract's 99% recall claim helps ROI conversations, but only if you can price the tier that includes it. Three-year TCO is a guess until Growth numbers surface.

Dropping a messy multi-doc PDF into Extract and getting typed JSON back with a citation and bounding box on every value is the feature that matters at 4pm when someone questions your numbers. You click the box, see the source line, move on. That's the whole point of a citation viewer, and Studio ships it free on every plan, unlike bolting a review UI onto an API-only tool like Textract.

Split's plain-language page-range logic beats writing regex against page breaks, and Classify's per-criterion confidence scores are genuinely useful for gating a pipeline before it hits a human queue.

Day-3 friction: credit-based pricing at $0.015 per credit after the 15,000 free tier means cost modeling requires actually running documents through first — no calculator, no published Growth tier pricing. Compared to a flat per-page rate, that's an extra spreadsheet you didn't want to build. Docs cover Parse/Split/Extract cleanly; Edit's vision-based field detection is less documented for edge cases.

This one's built for developers, not for someone opening a dashboard every morning, so my usual daily-polish questions get a little sideways. But the bones are there: 15,000 free credits to start, then $0.015 a credit, and a Studio workspace with a side-by-side citation viewer so you can actually check if the bounding box matches the number it pulled. That citation-on-every-value thing matters more than it sounds, because the alternative is trusting a black box with your invoice data.

The five-layer workflow, Parse, Split, Extract, Classify, Edit, reads clean on paper, and routing across 12+ models instead of making you pick one is a real convenience versus stitching together your own pipeline of, say, an OCR tool plus a separate LLM extraction layer. Competing against players like Textract-style IDP stacks, that consolidation pitch is the whole argument.

The tradeoff: no mobile story at all, no stated uptime or spinner behavior, and pricing opacity above Standard. Fine for an API-first buyer. Less fine if you wanted to just log in and see it work.

Three tells I look for: benchmark claims, named customers, real pricing. Reducto has two out of three. The 99% recall claim on micro1's LongExtractionBench is specific enough to check, and Harvey, Scale AI, Vanta as customers is a real signal — not vaporware logos.

What's missing: no public pricing past $0.015/credit on Standard. Growth and Enterprise are both 'contact sales,' which is category-normal but still opaque.

Competitive field is crowded — Textract, Azure Document Intelligence, Unstructured, LlamaParse all chase the same PDF-to-JSON problem. Reducto's pitch is consolidation: replace 4-5 vendors with one. That's the same story Unstructured told two years ago. Exit portability is fine — structured JSON output, standard SDKs, no lock-in trickery visible. Deep Extract and citation/bounding-box verification are genuinely differentiated features, not just repackaged OCR. SOC 2, HIPAA, air-gapped deployment options suggest enterprise-serious infrastructure, not a weekend API wrapper.