Google ships Lyria 3.5 in Flow Music

DeepMind claims advances in musicality, lyrics, vocals and creative control. No benchmark, no listening test, no price and no output-length figure accompanies the claim.

Google DeepMind has announced Lyria 3.5 in Google Flow Music, describing advances across musicality, lyrics, vocals and creative control. The announcement is dated July 29.

That is the honest extent of what we can report, and the reason is worth stating plainly. The DeepMind blog URL redirects to blog.google on a different host, and our fetcher does not follow cross-host redirects, so we did not retrieve the post. What we can say about the shape of the claim we can say from the announcement's own framing: it names four axes of improvement and attaches no measurement to any of them. Generative music releases rarely carry a benchmark, because there is no accepted one — no MMLU for melody, no standard listening test, no published preference study. So a version number and four adjectives is typically the whole of the public evidence.

We score it a 2 on that basis, capped as a vendor claim, and file it under CONNECTION as a software tool rather than under KNOWLEDGE where the automated pull put it. It is also eight days old, which is worth noting on a week where the fallback has repeatedly surfaced stale RSS items as though they were today's news. If Google publishes a listening study or Flow Music publishes a price, this becomes a gradeable story.

Source: Google DeepMind