Reducto funding hits $108M to feed AI messy docs

Reducto, a startup building the essential plumbing for AI to understand real-world documents, has raised a $75 million Series B led by Andreessen Horowitz (a16z). The new round brings the company’s total funding to $108 million and signals a massive bet on a less glamorous but critical piece of the AI puzzle: data ingestion.

While the industry obsesses over building bigger and better large language models, Reducto is focused on the messy, unstructured data those models need to be useful. Most of the world’s most valuable information is locked away in PDFs, spreadsheets, scans, and slides—formats that are notoriously difficult for AI to parse accurately. Reducto’s API acts as a universal translator, combining computer vision with vision-language models (VLMs) to convert this chaos into clean, LLM-ready data.

I’m thrilled to announce @reductoai’s $75M Series B led by @a16z, which brings our total funding to $108M.

Just five months after our Series A, we've surpassed 1 billion pages processed and grown our monthly volume 6x. We now process hundreds of millions of pages every month for… pic.twitter.com/6Lz1pzMatU

— Adit (@aditabrm) October 14, 2025

The company’s traction suggests it’s solving a very expensive problem. In the six months since its Series A, Reducto says its monthly processing volume has skyrocketed by 6x, having now processed over a billion pages. Its client list is a mix of the new AI guard, like legal AI startup Harvey, and the old guard, including a Fortune 10 company and a top-five global hedge fund.

An engineering leader at that fund said Reducto is "probably the only AI product that has actually worked for us," helping it parse documents with table complexity that was previously impossible.

The AI Gold Rush's Plumbing

Reducto isn’t just pulling text off a page. The platform offers a suite of tools to parse, split, extract, and even edit documents programmatically. It’s a key ingredient for building the next wave of “agentic” AI applications—systems that can read a report, extract key figures, and then act on that information without human intervention. The company is already expanding its “Agentic OCR” framework to improve chart extraction and review, pushing for what it calls "near-perfect accuracy."

With the new capital, Reducto plans to accelerate its model research and scale its go-to-market efforts. Crucially, it’s also introducing a more flexible pricing structure aimed at early-stage startups and researchers. This move positions Reducto not just as an enterprise solution but as the default ingestion layer for anyone building with AI, from a two-person startup to a Fortune 500 giant.

As a16z and other investors like Benchmark and Y Combinator pour money into the company, the message is clear: the AI revolution won’t be built on models alone. It will be built on the tools that can reliably connect those models to the messy, complicated, and overwhelmingly human data that runs the world.