Tier 2 Restricted open-weight
What the publisher claims: open, open research, responsible AI
The one-liner: Open weights wrapped in a purpose-built responsible-AI licence.
The six questions
weights downloadable, ungated
custom RAIL terms restrict use rather than granting unrestricted rights
deployable in standard tooling
dataset composition, training details and evals documented in the paper and model card; corpus not published
Limited to moderate.
BigScience was a 1,000-plus-author, multi-institution workshop whose model card, dataset curation, and RAIL licence were argued in public across its GitHub and Hugging Face spaces, so outside researchers steered the release rather than merely consuming it.
License
BigScience RAIL License v1.0 — license source ↗
- custom Responsible AI License with a prohibited-use list (discrimination, medical advice, law-enforcement and justice-administration uses barred)
- permissive grant modified from Apache 2.0
What is actually released
- weights (176B and smaller)
- model card with full dataset composition
- technical report
- training code
- evaluation information; training dataset itself not released
Why it matters
The origin point of the RAIL licence stack that later open-weight releases copied, including the pattern of attaching use restrictions to otherwise permissive terms.
Sources
- https://huggingface.co/bigscience/bloom-560m/blob/main/LICENSE
- https://huggingface.co/api/models/bigscience/bloom-560m
Derived from the openness dataset behind But Is It Open? — snapshot 2026-11-07 ↗. Scores are qualitative readings of the cited evidence, not verdicts; tier measures openness, not intelligence.