Tier 3 Practical open-weight
What the publisher claims: open-source, open-weight
The one-liner: Permissively licensed embedding family that is the de facto default for retrieval.
The six questions
weights downloadable
MIT or Apache 2.0 depending on checkpoint
light and trivially deployable
model cards and evals only; corpus withheld
Limited to moderate.
BAAI maintains active repos with merged outside PRs, but direction and releases stay corporate.
License
MIT for bge-m3 and the v1.5 encoders; Apache 2.0 for bge-reranker-v2-m3 - verify per checkpoint OSI-approved — license source ↗
What is actually released
- weights
- ONNX builds
- sentence-transformers configs
- model cards with training data description
- MTEB evaluation results
Why it matters
Embeddings carry the same openness questions as generation models, but the licence answer here is unusually clean.
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.