gemini-embedding-2-preview now routes to the GA embedding model
TL;DR
On November 6, 2026 Google deprecated gemini-embedding-2-preview. Requests to that model string now route to gemini-embedding-2 and are billed at Gemini Embedding 2 rates. Google says the schema, price, and embedding space match, so existing vectors do not need to be rebuilt.
- Status
- Live
- Released
- Plans
- Gemini API
- Source
- —
What changed
The Gemini API release notes for November 6, 2026 say gemini-embedding-2-preview is deprecated. From that date, API requests that still name the preview model are routed to gemini-embedding-2 and billed at Gemini Embedding 2 rates.
Google says gemini-embedding-2 uses the same API schema, the same pricing, and the same embedding space. The note says no other code changes are required, and vectors do not need to be re-indexed. You should still change the model string to gemini-embedding-2.
Why it matters
A preview model string left in a job will keep returning embeddings, but it is no longer the model you think you are calling. Because Google says the embedding space is unchanged, old vectors stay comparable with new ones. That is the claim to test if you mix old and new indexes in one search.
Update the string anyway. A later removal of the preview alias would break the call, and bills should name the GA model you actually intend to use.
Frequently asked questions
Do I have to rebuild my vector index?
Google says no. The preview alias routes to gemini-embedding-2, which it says uses the same embedding space, schema, and pricing.
What should I change in code?
Set the model string to gemini-embedding-2.