- Home
- Training data
AI training data for every modality.
Rights-cleared data across audio, video, images, text, code, reasoning, human feedback, robotics, and business operations. Licensed off the shelf where we have it, collected to your spec where we don't.
Audio & Speech
Conversational, accented, expressive, and low-resource speech, plus long-form audio.
Video
Long-form, instructional, egocentric, and action video with dense captions.
Images & Vision
Captioned images, documents, charts, products, and geospatial imagery.
Text & Documents
Books, journalism, scientific, legal, and financial text, in many languages.
Code & Software
Permissive source, repositories, pull requests, execution traces, and SQL.
Reasoning & Evaluation
Chain-of-thought, math, verifiable problems, agent trajectories, and private evals.
Human Feedback & Safety
Preference pairs, demonstrations, red-teaming, moderation, and factuality labels.
Robotics & Sensor
Teleoperation, manipulation, driving, LiDAR, IMU, and drone data.
Business Operations Data
Anonymized messaging, documents, CRM, tickets, and code from real companies.
From sample to production.
Most teams test fit on a sample, run a pilot, then scale. Nothing is committed until the data has passed your own evaluation.
Brief
Tell us the model, stage, and constraints. We reply with the relevant catalogue and options.
Sample
A representative slice and the full datasheet, free, to run through your evaluation.
Pilot
A scoped delivery against acceptance criteria you set, before any large commitment.
Production
Full delivery to your storage, on one schedule or in rolling batches.
Refresh
New data, new languages, or new labels as your training needs change.
What every datasheet covers.
The documentation your research and legal teams will ask for, delivered with every sample and every order.
- Source and collection methodWhere the data came from and how it was gathered.
- Rights and consentThe license chain and consent basis, with an audit trail.
- CompositionLanguages, domains, demographics, and how they are balanced.
- AnnotationThe label schema, guidelines, and who applied them.
- Quality resultsHow quality was measured on this data, and what was found.
- Cleaning and deduplicationFiltering, near-duplicate removal, and contamination checks.
- Known limitationsGaps, biases, and uses the data is not suited for.
- VersioningWhat changed between deliveries, so results stay reproducible.
Quality you can check.
Measured, not asserted
Label agreement and audit samples are measured on your delivery and reported in the datasheet.
Gold references
Annotators are calibrated against expert-labeled reference sets before and during production.
Your acceptance criteria
Quality thresholds are written into your order, and deliveries are checked against them.
Ready for procurement.
Built to get through legal, security, and finance review at large labs and enterprises.
NDA first
We share the detailed catalogue under your NDA or ours.
Data protection
A Data Processing Addendum for any personal data, with privacy handling documented per dataset.
Security review
We complete security questionnaires and deliver straight into your own storage.
Clear contracts
A license agreement with ModelCore LLC and an order form per delivery. How licensing works
Request the catalogue.
Tell us what you're building and we send the relevant catalogue, with volumes, specs, and sample availability. Some datasets are shared only under NDA.
- Free samplesA representative slice and datasheet before any commitment.
- Matched to your useWe send what fits your model and stage, not a generic list.
- Reply within one business dayFrom our data team, not an autoresponder.