Google released EmbeddingGemma 2 under the Apache 2.0 licence. This decision addresses a specific concern regarding proprietary hosted models. Most embedding applications require calculating thousands of vectors and storing them for later comparison. If a vendor stops offering a closed model, users must pay to recalculate millions of stored vectors. OpenAI previously covered such costs in April 2024, but that cannot be relied upon from every provider.
The open licence allows developers to run the open weights version locally if a provider discontinues hosting. This offers a safety net without requiring users to host the model themselves. It ensures long-term access to the model architecture regardless of commercial decisions.
* Apache 2.0 licence permits local deployment
* Avoids vendor lock-in for vector storage costs
* Maintains access if Google stops hosting the service
