Simon Willison argues that open weights offer an option if a hosted embedding model is discontinued: an operator could run the weights themselves or use another provider. He notes that embedding applications may calculate thousands or millions of vectors and store them for later comparison, making re-embedding a potential cost when models change. The source is commentary, not a full release specification; it does not establish model capabilities, hosting availability, or migration terms. Source: https://simonwillison.net/2026/Oct/6/hn-49983751
Raw SignalOct 6, 20261 min read1 source
Google s embeddinggemma 2 open embedding model and apache 2 0 licensing
In one minute
Google’s EmbeddingGemma 2 is described in the supplied source as being under the Apache 2.0 license.