Tier 3 Practical open-weight
What the publisher claims: open-source, open-weight
The one-liner: Permissively licensed industrial open weights with a published training-content summary.
The six questions
weights downloadable, ungated
MIT
standard tooling, multiple quantizations
training-content summary and tech report published; corpus not released
Limited to moderate.
Ant Group's inclusionAI org publishes Ling weights and training code under MIT with a 'Get Involved' framing and merged outside PRs to its frameworks, yet Ant Group alone decides what ships and when.
License
MIT (HF license:mit tag on Ling-2.6-flash, Ling-2.6-1T and their fp8/int4 variants) OSI-approved — license source ↗
What is actually released
- weights (Ling-2.6-flash 104B MoE with 7.4B active, Ling-2.6-1T)
- fp8 and int4 variants
- custom modeling code
- tech report arXiv 2606.15079
- public training-content summary document
- evaluation results
Why it matters
A useful reference point for how a large non-Western lab describes its training data without opening the corpus.
Sources
- https://huggingface.co/inclusionAI/Ling-2.6-flash
- https://huggingface.co/api/models/inclusionAI/Ling-2.6-flash
- https://huggingface.co/api/models/inclusionAI/Ling-2.6-1T
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.