Thesis
Apache 2.0 licensing for EmbeddingGemma 2 gives embedding-system operators an alternative to relying exclusively on a hosted proprietary model.
Observed facts
Simon Willison says EmbeddingGemma 2 is under the Apache 2.0 license.
1He says embedding applications often calculate thousands or millions of vectors and store them for later comparison.
1He describes the possibility that a vendor may stop offering a model, leaving users to recalculate stored vectors.
1
Mechanism
When a model changes or disappears, vectors produced by different models may require recalculation to maintain a consistent index. Open weights can preserve an operator’s option to run that model independently or seek another provider.
Why now
The release makes licensing and fallback options salient for teams choosing an embedding dependency.
Implications
Teams with large stored indexes should account for model continuity and re-embedding costs when choosing a provider.
Open weights may offer a fallback path, while hosted inference can still be the preferred operating model.
Evidence boundary
This record relies only on the linked commentary and does not independently establish those operational details.
Watch conditions
Whether providers offer stable versioning or migration support will determine how much practical protection open weights provide.
Check the official release materials for model specifications, supported deployment paths, and terms of use.