NuMind blog · May 2026
NuMind released NuExtract3, a 4B open-source VLM that unifies structured extraction and OCR, adds reasoning via reinforcement learning, and claims better performance than similarly sized models.
Company research
Public company, workplace, funding, and market signals
Updated Jul 29, 2026
Overview
NuMind is a YC S22 AI startup that builds specialized multimodal LLMs for document extraction and OCR, sold through a SaaS platform and enterprise/private deployment options. Its current commercial/product line centers on NuExtract, a document-extraction model and platform suite.
Primary product
NuExtract
Founded
2022
Headquarters
Cambridge, Massachusetts, United States
Team size
10-20
Work style
Hybrid
Industry
Software Development
Sub-industry
Natural language processing (NLP) / document extraction AI
Offices
Cambridge, Massachusetts, United States
Paris, France
Station F, Paris 13, Paris, France
Business model
SaaS
enterprise deployments
usage-based API
open-source model releases
Funding
Stage
Seed
Total raised
$3.6M
Latest round
Seed · Mar 2023
Latest amount
$3M
Mar 2023 · Flybridge
Convertible Note / early round
Sep 2022
Jun 2022 · Pioneer Fund
People & investors
Work & culture
Research-driven and product-focused, with emphasis on rigorous ML research, fast product innovation, autonomy, collaboration, and using AI tools in daily development.
Work style
Hybrid
Compensation
A public fullstack job posting listed €45k–75k base salary plus stock options, with CDI and RTT; no broader company-wide compensation policy was found.
Benefits
Stock options
Permanent contract (CDI)
RTT
Partially remote work
Dedicated office at Station F
Product & market
Pricing
Usage-based API pricing at $5 per million tokens, with the web UI free; enterprise/private deployment available.
Differentiators
- —Specialized for document extraction and OCR rather than general chat
- —Structured extraction to JSON plus content extraction to Markdown
- —Open-source models under MIT or Apache 2.0 licenses
- —Private/on-prem deployment options for data confidentiality
- —Explicit null output when information is absent, reducing hallucinations
- —Multilingual and infinite-context capabilities
- —Task templates plus examples for extraction workflows
Technology
multimodal LLMs
OCR
JSON extraction
Scala
Laminar
Tapir
Ox
Hugging Face
Python SDK
Customers
AXA
medical coding workflows
legal report review workflows
user feedback analysis workflows
recruitment / resume-matching workflows
Competitors
Capacity
Machinify
InsightFinder AI
LightOn
Reducto
Upstage
Aleph Alpha
Scale & contact
Estimated revenue
under $1M
Estimated monthly visits
3.4K
Traffic estimate as of Jul 2026
Latest news
LinkedIn · Apr 2026
NuMind said it is working with AXA on document extraction and reported large accuracy gains on Japanese health-claim documents.
NuMind blog · Jul 2025
NuMind launched the NuExtract Platform with API and web UI, priced at $5 per million tokens for API usage and available for private deployment.
NuMind blog · Jul 2025
NuMind released NuExtract 2.0, a specialized document-extraction model line claiming strong benchmark performance against frontier LLMs.
NuMind blog · Oct 2024
NuMind released NuExtract 1.5, adding multilingual support and an infinite-context continuation mechanism for structured extraction.
Jobs and careers at NuMind
0 jobs at NuMind
No open positions
Check back later for new openings