Gary Illyes from Google said that despite what people are saying, he did not confirm Google uses MUVERA - Multi-Vector Retrieval via Fixed Dimensional Encodings. He said that Google might use something like it in Google Search but he did not confirm that Google actually uses it today in Google Search.
Gary Illyes said this in a LinkedIn post, when that post said:
Gary Illyes, while answering to questions at the recent Search Central Live Deep Dive in Asia Pacific Event 2025, said that Google is using something like MUVERA in production, though they might refer to it by a different internal name. That’s huge.
Gary replied:
No, I didn't confirm we're using it. All I said was that we might have something like it.
What is MUVERA? MUVERA is a retrieval algorithm used in information retrieval (like search engines). Its purpose is to make multi-vector search faster and more efficient. Multi-vector search, while more accurate than single-vector search, is also more computationally expensive. MUVERA addresses this by converting complex multi-vector data into a single, simpler representation called a "fixed dimensional encoding" (FDE), which can then be searched using highly optimized single-vector algorithms. The original, more complex multi-vector similarity is then used to re-rank the top results for accuracy.
MUVERA could help Google understand your queries in a more nuanced way, especially for complex or conversational searches. Instead of treating your query as a simple collection of keywords, it would break it down into multiple "meaning blocks." The multi-vector approach allows MUVERA to find documents that are a better semantic match to your query. This means it can find relevant pages even if they don't use your exact wording.
It is cool and you may have fun reading about the details of MUVERA on the Google blog but just because it is on the Google research blog, it does not mean Google Search uses it right now.
Hat tip to Gagan Ghorta:
🆕 Googler Gary Illyes about MUVERA and Google using it (commenting about the previous thing that he said (quote tweet))
— Gagan Ghotra (@gaganghotra_) August 4, 2025
"No, I didn't confirm we're using it. All I said was that we might have something like it".
via LinkedIn comment https://t.co/nlqay4GpQD https://t.co/BacJhhCFxY pic.twitter.com/HP7lBY43XX
Forum discussion at LinkedIn.
