brain/
sourceartificial-intelligence

Lyria 3.5 lands in Gemini app and API (global)

Sep 4 Google issuer post plus official GeminiApp X: Lyria 3.5, already in Flow Music since Jul 29, is now in the Gemini app, Gemini API, AI Studio, and Google Vids; all users globally on web and mobile.

Source

Lyria 3.5 lands in Gemini app and API (global)

Generated by Grok Bot research on 2026-09-04. WebSearch + fetch ladder + native X. Treat as raw material — review before promoting.

Summary

After 10:00 AM ET on 2026-09-04, the only high-signal frontier-lab original on the weekday X pass was Google product, not a new foundation-model card. Verified @GeminiApp said Lyria 3.5 is in Gemini, then all users globally on the web and in the app. The matching issuer page, dated Sep 04, 2026, is Create your best tracks yet with Lyria 3.5 in Gemini. That post also names the Gemini API, Google Flow Music, Google AI Studio, and Google Vids. The model itself is older: Google launched Lyria 3.5 on Jul 29, 2026 in Flow Music. Do not file Sep 4 as a new-model day.

Findings

Sep 4 access expansion, not a new model

The Sep 04 Keyword post states: “Lyria 3.5, our best-sounding music generation model, is now available in the Gemini app and the Gemini API,” with “more expressive vocals and richer musical arrangements” and “higher fidelity.” In the Gemini app it lists: select or describe genre and choose vocal or instrumental; new templates; flexibility to choose short or longer tracks. It then says Lyria 3.5 “is available to all users globally on the web and in the mobile app,” and “also available for artists and AI creatives in Google Flow Music, and for developers and technologists through Google AI Studio, and Google Vids.”

Those sentences match the two official X posts the same afternoon (16:02 and 16:22 UTC). The follow-up’s t.co target resolved to gemini.google.com/music. Neither hydrated post returned a media attachment, so this clip does not set x_video.

The Jul 29 issuer post Introducing Lyria 3.5 in Google Flow Music is the model-intro date: musicality, lyrics, vocals/pronunciation, and tempo/duration control, rolling out that day in Flow Music. Sep 4 is Gemini-surface + API distribution of that already-named SKU.

What the Gemini marketing page still says

Fetched gemini.google/overview/music-generation/ still titles the product Lyria 3, and still splits SKUs as Lyria 3 (free, tracks up to 30 seconds) vs Lyria 3 Pro (higher limits with Google AI Plus/Pro/Ultra; tracks up to 3 minutes). FAQ text still says to pick “Fast” for Lyria 3 / 30 seconds and “Thinking” or “Pro” for Lyria 3 Pro / up to 3 minutes. It also restates 18+ availability wherever the Gemini app is available, and SynthID watermarking. That page does not name Lyria 3.5. Treat it as lagging marketing HTML, not as a contradiction of the dated Sep 4 blog.

The Feb 18, 2026 Gemini-app launch post Use Lyria 3 to create music tracks in the Gemini app is prior: 30-second tracks, 18+, listed languages, paid-plan higher limits. Its related-story chip now points at the Sep 4 Lyria 3.5 Gemini post.

Quiet labs after the 10am clip

get_users_posts with start_time 2026-09-04T14:00:00Z (after today’s AI-thread X source) returned zero original posts from @karpathy, @OpenAI, @AnthropicAI, @GoogleDeepMind, @sama, @Zai_org, and @opencode. News-cluster summaries (Astra recaps, Amodei-meme chatter) were used only as pointers and are not cited as grain.

Contradictions and open questions

  • SKU rename vs page lag: Sep 4 blog/X say Lyria 3.5 for all Gemini users globally, short or longer tracks. The overview page still sells Lyria 3 / Lyria 3 Pro with 30s vs 3-minute and paid-plan language. Do not collapse 3.5 into 3 Pro, and do not invent a 3.5 length cap.
  • API grain missing: The blog names the Gemini API, AI Studio, and Vids but does not publish a model id, price, or rate limit in the fetched text. Do not invent those.
  • Flow Music vs Gemini: Jul 29 launched 3.5 in Flow Music; Sep 4 adds Gemini app/API and restates Flow Music. Whether Flow Music and Gemini share one stack is not evidenced here.
  • “Best-sounding” is Google marketing, not an independent eval.

Provenance

Method: Grok Bot / WebSearch / fetch ladder / native X Generated: 2026-09-04 Rounds: 2 of 3 (early-exit — issuer blog + X posts hydrated; remaining uncertainty is API ids/pricing, not worth a third round) URLs fetched: 5 successful, 2 timed out Fetch notes: WebFetch timeout on https://gemini.google.com/music (t.co expanded via curl Location to that URL) and on https://deepmind.google/models/lyria. Issuer blogs and the music-overview page fetched cleanly. X via official MCP (search_news pointers only; get_users_by_usernames + get_users_posts + get_posts_by_ids). Skipped search_posts_all. No media on the two GeminiApp posts.

Web sources:

X sources:

Grokipedia:

  • not used
Referenced by