If you printed and stacked the pages @reductoai has processed this year, the stack would reach space multiple times over We just wrapped an insane Q2 but some highlights that stand out: - 🖥️ We made a lot of progress on core model performance and grew the product surface. Reducto is now faster and more accurate across the board - 📈 Our GTM team is crushing it, and the team delivered 350% to quota capacity this quarter - 🖥️ (pt 2) Our new products like Deep Extract are setting a new standard for the industry. In @micro1_ai's benchmark there was literally no solution that came close. - 💰We broke our record for largest new contract twice - 🏢 We're now 50% over intended capacity for our office, and are moving to a beautiful new space soon --- The day to day of startup life is always chaotic with plenty of ups and downs, but every time I pause to think about the bigger picture of our journey it's genuinely hard to believe. Revenue growth is still around 8x YoY, and we're just now ramping our capacity. Our sales team in Q3 will be 4x the size of what it was in Q2. Even more importantly, we secured a massive cluster of GPUs and there's a lot that our product research teams are working on that I think will be a step change for our customers. As always, could not be more grateful for our customers and for the team that makes all of that possible. More from us soon 🔮

REDUCTO ON THE ROAD 🗣️‼️

Today we're publishing LongExtractBench, a benchmark commissioned by @reductoai and independently validated by micro1. We evaluated seven production document extraction systems across the same 225 complex enterprise documents. The benchmark was intentionally difficult: documents averaged 358 pages and contained roughly 88,700 ground-truth fields each. Every system was evaluated using the configuration documented in the benchmark methodology. Key findings: • Reducto Deep Extract was the only system to successfully complete all 225 documents. • Direct frontier LLM baselines achieved substantially lower completion rates on long, complex documents. • In this benchmark, dedicated extraction platforms achieved higher completion rates than the direct frontier LLM baselines. • Recall was the clearest differentiator. Precision remained high across systems, but recall ranged from 33.8% to 99.6%, highlighting which systems consistently captured the information contained in long, complex documents. The full report includes the benchmark methodology, limitations, and reproducibility resources. Check out the report and results in the comments below.

We see this launch as an important way to highlight the silent failures that exist outside of marketing content, but the best evaluation will always be on your own data. If your company uses another extraction service, we’re offering up to $1000 in credits for you to compare Deep Extract directly. You can sign up at: reducto.ai/try-deep-extract

Our agent harness, Deep Extract, was the only solution that was both complete and accurate. On the unfiltered test set, Reducto: - Completed 225/225 documents - Achieved precision and recall above 99% - With 0 failures Supporting real world use cases requires more than just impressive accuracy. Enterprises need their pipelines to be accurate and reliable, and this benchmark highlights all of the work our team has put into making Deep Extract production ready.

The dividing line across solutions was the tradeoff between precision and recall. Frontier models performed well on precision for completed documents, achieving 90% scores that surpassed some dedicated document extraction services. On the other hand, all models struggled with recall. Even on the easier subset of common-completed documents, models fell below the 80% mark and repeatedly dropped rows and important values.

micro1’s blog: micro1.ai/benchmark/long-ext… Reducto’s analysis: reducto.ai/blog/reducto-lead… – We commissioned micro1 to create a benchmark that tests challenging structured extraction tasks from REAL documents. The dataset tests 225 documents with an average of 358 pages and roughly 88.7K ground truth fields each. Reducto did not select the tested documents or modify the ground truth in any capacity. The dataset records failures but reports accuracy on the completed set for each provider to avoid conflating the two metrics.

Many companies are #1 in a benchmark they crafted. We worked with @micro1 to create an independently audited benchmark to measure document extraction performance with long documents. The results of LongExtractBench show the nuances companies are likely to find in the real world. micro1 tested frontier models with max reasoning and document processing platforms with their strongest configurations, and found notable precision/recall and completion tradeoffs across most. Reducto’s Deep Extract leads the industry by a wide margin. 🧵

Today we're launching @SpellbookLegal's biggest thing yet: Autonomous Contract Management It’s the first end-to-end AI infrastructure for contracts The world is speeding up. We are in one of the biggest investment cycles in decades. Behind every rocket launch, FIFA game, and datacenter, lies a web of hundreds of agreements. Agreements are the invisible threads that allow us to work together. We have infrastructure for finance (Stripe, Ramp), eCommerce (Shopify) and many business functions. But agreement infrastructure is lacking. This creates a painful bottleneck on our ability to work together. Online purchases take milliseconds. But agreements still take weeks. CLMs were supposed to be the answer, but were designed in a pre-AI era. AI fundamentally changed how computers can accelerate agreements. Spellbook is the most used AI contract review tool in the world, with ~5,000 customers in 80 countries Now we are expanding to deliver the first end-to-end, AI native stack for contracts. From the moment a deal lands in your inbox, to the day it renews years later, Spellbook’s AI supports teams every step of the way, across all business teams. It runs 24/7. While you sleep, it's reading the deals that came in overnight, flagging the parts that actually need a lawyer, and clearing the busywork that used to kill mornings. Nothing gets handed off between systems or slips after signature, and the intelligence stays with you for the life of the contract. AI for lawyers is great. There are 20 million lawyers in the world, and many are our users. But there are billions who touch contracts. We're excited to help everyone move faster and do more of what they love, by building the best AI-powered contract infrastructure in the world. Get early access: spellbook.com/acm

Our cookbooks got a new look 👨‍🍳 we now have recipes to help you: > Turn bank statement PDFs and scans into a clean, typed list of every transaction ready to load into a ledger or audit trail. > Pull demographics, insurance, and medication history from patient intake forms into structured JSON with a source citation on every field. > Read every strikethrough, underline, and annotation in a redlined contract as structured tags so you can surface all revisions in code instead of page-by-page.

If you ever felt like you don't have enough Reducto in your life, we're fixing that this week at @aiDotEngineer ☺️ We have *5* events this week, a few thousand dollars worth of swag to give out, and prizes at our booth (P8) Come say hi!

thx for coming to the s26 kickoff @aditabrm and sharing lessons on enterprise sales for AI products

every procurement conversation should end with giving a 1-5 star rating kinda like an uber ride

i'm hosting a small dinner with the best founders in difficult vertical ai spaces on 7/1- finance, healthcare, legal, and more. it'll be great to share notes on building trust, enterprise sales, and more! dm me or comment for the invite

Fun fact: even though we didn't work on document processing until 2024, our average daily pages processed from this incorporation date to now is still well into the millions

Exactly 1000 days since @raunakdoesdev and I decided to start a company together, with zero clue what we would build A lot of memories (and pages processed) since

Realized mid cut that my barber messed up because he kept getting more stressed as he tried to fix it

this sales draft is our world cup

You don't really need Fable. Opus with better inputs outperforms Fable on Surge's GDP.pdf benchmark. It also leads to fewer reasoning tokens, lower latency, and better cost at scale.