The Reducto Roundup

No dark-mode styles — this email renders the same in a dark client.

Town’s $108M Series B, behind the build with Drata, and the latest in r-1.


https://reducto.ai/?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=header_home

Happy Tuesday!
We’re back with another Reducto Roundup. After a busy couple of weeks in New York and Tulum
[https://www.linkedin.com/posts/aditabraham_we-just-got-the-footage-from-our-offsite-ugcPost-7518041177011470336-ScoT/?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=tulum_video],
we’ve got plenty to share: the latest in AI, what we’ve been building, and a few things worth trying yourself.

What’s happening in AI
• New Flash models from MiniMax and DeepSeek. MiniMax’s M3.1 Flash [https://www.minimax.io/blog/minimax-m31f] arrived October 19,
following DeepSeek’s October 17 update [https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017]. Both teams report better
performance on coding and agent tasks, with faster completion times.
• Replit switches its default to OpenAI [https://techcrunch.com/2026/10/17/replit-is-no-longer-working-with-anthropic-ceo-says/].
Users can still choose Claude.
• Muse Wonder handles support better than design. In Claire Vo’s hands-on review
[https://www.chatprd.ai/how-i-ai/meta-muse-wonder-review], Meta’s model stood out on billing tickets and inbox triage, but
struggled to finish polished, working interfaces.
• Zocdoc buys Sohar Health. The acquisition
[https://www.zocdoc.com/about/news/zocdoc-acquires-sohar-health-to-make-insurance-easier-to-navigate-for-patients-and-providers/]
builds on a partnership already helping patients verify insurance coverage. Sohar automates eligibility and benefits checks.

Companies to watch
Congratulations to Town on its $108 million Series B [https://www.town.com/press/series-b], at a $1 billion valuation! Town brings
AI assistants into the tools teams already use, helping with email, meetings, and projects. So much of enterprise work lives in
tricky document uploads. We’re excited to continue supporting Town with Reducto as the team grows and brings AI to more of that
work.

Behind the build with Drata



A security questionnaire could mean up to 30 minutes of manual prep before Drata’s AI could start answering it.
Now, Drata uses Reducto to extract questions from hundreds of questionnaires each month and write approved answers back into the
original files, so the process can start without waiting for someone to prepare the file.
Hear how they built it from Lior Solomon, VP of Engineering, AI and Data, and Julien Melissas, Engineering Manager, at Drata.
See Drata’s questionnaire workflow →
[https://reducto.ai/blog/reducto-drata-customer-story?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=drata_case_study]

Inside r-1: how it reads, how we test it
A chart can mean something different when you read the paragraph beside it. Our parsing model, r-1
[https://reducto.ai/blog/parse-r-1-model?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=r1_model],
uses that context to make sense of figures, tables, and handwriting. Here’s a look at how it works and how we check the results.

How r-1 reads figures →
[https://x.com/raunakdoesdev/status/2109678596834873756?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=r1_figures_video]

Our ML engineers Ani and Chuyi show how the surrounding page helps r-1 interpret a figure beyond what’s inside the image.

Doc Benchmarking 101 →
[https://x.com/adelwu_/status/2107207290717933917?s=20&utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=benchmarking_video]

Archana from product and Adel from marketing explain how to separate cosmetic differences from meaningful errors, compare
equivalent setups, and check your LLM judge’s decisions.

One lesson from our evaluations: a judge sometimes preferred “This figure is a graph” over a description that actually explained
its contents. Correct doesn’t always mean useful. Our engineering blog explores what makes a good figure summary
[https://reducto.ai/engineering/posts/what-makes-a-good-figure-summary/?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=figure_summaries_article]
and why those missing details matter.
For an independent evaluation, Verona’s K1Bench
[https://verona.so/blog/k1bench?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=k1bench]
ranked r-1 first across 196 pages from 20 real-world K-1 packets: 85.2% of pages had no material error, 9.2 percentage points
ahead of the next parser.

Everyone’s an engineer



Our team deployed nearly 200 apps in just two months, from customer demos to internal tools.
Our engineer Abhi Arya shares how we built the infrastructure that made this possible, simplifying authentication, access, and
deployment so anyone on the team can build.
How we made everyone an engineer →
[https://reducto.ai/engineering/posts/everyone-an-engineer/?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=everyone_an_engineer]

Upcoming Events
New York & San Francisco
Data Outpost



Join Josh Nkoy, our Founding Solutions Engineer, and Palak Agarwal, our DevRel Lead, in San Francisco on November 4, from 2:30 to
4:30 p.m. PT.
Bring PDFs, scans, or images. We’ll use Reducto’s MCP to build a searchable archive and a small exhibition, from annotated poems
to a season’s worth of box scores. You’ll leave with a working document pipeline.
Register for Data Outpost →
[https://www.dataoutpost.ai/?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=data_outpost#from-unstructured-documents-to-agent-ready-data]

Non-Slop Hackathon with Pangram and Nous Research

https://luma.com/enrexhqq?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=non_slop_hackathon

We’re sponsoring a day of building with Pangram and Nous Research in Brooklyn on Sunday, November 22. Every participant gets
$1,000 in Reducto credits, with a $1,000 cash prize from us up for grabs.
Sign up for the hackathon →
[https://luma.com/enrexhqq?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=non_slop_hackathon]

Something to try: SOC 2 review
Working through a vendor’s SOC 2 report? Our cookbook shows how to turn auditor exceptions and control results into a review
queue, with each finding linked to its source.
It also checks the report’s scope against a reviewer’s checklist and flags untested procedures, so reviewers can see what needs
attention and trace each finding back to the evidence.
Read the SOC 2 cookbook →
[https://reducto.ai/cookbooks/soc2-exception-register?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=soc2_cookbook]
You can start by uploading a report to Studio
[https://studio.reducto.ai/?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=try_studio].
P.S. Reviewing SOC 2 reports? Take a look at how Vanta approaches evidence checks
[https://reducto.ai/blog/vanta-reducto-case-study?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=vanta_case_study].



Until next time,
The Reducto team

https://reducto.ai/?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=footer_home

Reducto is the complete agentic document platform for leading AI teams needing performance at enterprise scale.
Reducto, Inc. © 2026

https://x.com/reductoai?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=footer_xhttps://www.linkedin.com/company/reducto-ai/?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=footer_linkedinhttps://github.com/reductoai?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=footer_githubhttps://huggingface.co/reducto?utm_source=reducto_roundup&utm_medium=email&utm_campaign=reducto_roundup_2_37&utm_content=footer_huggingface

You are receiving this email because you opted in to receive updates from Reducto.

Reducto, 717 California St, San Francisco, CA 94115
[https://www.google.com/maps/search/717+California+St,+San+Francisco,+CA+94115]
Unsubscribe [https://app.loops.so/unsubscribe/cmvh5ochm0t7101xd82p9aris/0d1e76f3bbcd526d0e84db04125daf218791a76adca369b82b62704cc967841375052b18937c119a4450ecd7560855622d6bd7530bea55c7719c7c11]

Accession

nse/reducto/2026-10-20/4600

From

Vibha at Reducto · srivibha@hey.reducto.ai

Sent

20 October 2026 · 20:00 UTC