Tier 3 Practical open-weight
What the publisher claims: open, open data
The one-liner: Code model whose training data is genuinely open, including the opt-out mechanism.
The six questions
weights and the training dataset downloadable
commercial use permitted subject to use-based restrictions
standard Transformers, vLLM and local serving
dataset, opt-out handling, code and evals published
Moderate for research replication, not exact reproduction.
BigCode is a multistakeholder community (HF, ServiceNow, Red Hat and others) with public working groups and a governed release process.
License
BigCode OpenRAIL-M — license source ↗
- use-based restrictions with permitted-use and prohibited-use lists
- commercial use allowed but conditioned on use restrictions
What is actually released
- weights (3B, 7B, 15B)
- The Stack v2 training data published with opt-out requests honoured
- per-language opt-out mechanism and governance documentation
- training and fine-tuning code
- tokenizer
- evaluation results
Why it matters
The strongest example of consent-aware open training data, and a direct answer to the claim that code corpora cannot be published lawfully.
Sources
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.