Always published
- Papers and technical reports.
- Evaluation protocols and results, including failures.
- Model cards, for every model, whatever its access tier.
- The list of data sources and the licence of each.
Responsibility
Release policy
Three tiers, stated in advance, so nobody has to guess what we will share.
Published case by case
- Model weights.
- Datasets and full data recipes.
- Code for training and inference.
Each release states which of these it includes and why. Where we hold something back, we say what and give the reason: misuse risk, a partner's data terms, or patient safety.
Commercial
- Product code and infrastructure, owned by OneThought AI.
- Client data, which belongs to the client and is never used to train foundation models without consent.
Partner data
Data contributed by partners is used under the terms agreed with each partner. Those terms are summarised in the release that uses the data.
Version 1.0 · 23 October 2026