Lyria 3.5 brings full-length AI songs to the Gemini app and API

Google's Lyria 3.5 music model is now in the Gemini app and the Gemini API, with tracks up to three minutes long.

Google announced the Gemini app and API rollout in a blog post dated September 4, 2026. The model itself is older than that reporting. Lyria 3.5 first shipped in Google Flow Music on July 29, 2026, and the Gemini API changelog marked lyria-3.5 generally available on September 3, 2026, though one write-up below describes it as a preview model. Independent write-ups of the cross-ecosystem rollout by Ali Farhat, at scalevise.com on September 9, 2026 and on dev.to on October 31, 2026, both trail the model's release.

In the Gemini app, users can select or describe a genre, choose vocal or instrumental output, start from a template gallery, and pick short or longer tracks. Google's Gemini music page lists a maximum length of three minutes and says each track arrives with cover art made by Nano Banana. The API documentation instead describes Lyria 3.5 output as a couple of minutes, controllable through the prompt.

Lyria 3.5 is Google's flagship music generation model, built for full-length songs with verses, choruses, and bridges. It renders 44.1 kHz stereo audio from text prompts or from images. Its predecessor, Lyria 3, launched on March 25, 2026, as two models: lyria-3-clip-preview for 30-second clips and lyria-3-pro-preview for full-length songs, both at 48 kHz stereo.

The model page lists an input token limit of 131,072 and outputs of MP3 audio plus lyrics text, with caching, function calling, image generation, and structured outputs unsupported. The DeepMind model card says Lyria 3.5 uses latent diffusion over temporal audio latents, was trained on Google TPUs with JAX and ML Pathways, and was evaluated against Lyria 2. MLQ reports the API blocks prompts asking for a specific artist's voice or copyrighted lyrics, generates a track in a single turn with no iterative editing, and embeds an inaudible SynthID watermark in every track.

On the Gemini API paid tier, Lyria 3.5 costs $0.08 per full song, and the pricing page lists no free tier; the 30-second Lyria 3 Clip is $0.04. Lyria 3.5 is available in the Gemini app globally on web and mobile, in Google Flow Music, Google AI Studio, and Google Vids. Gemini app music generation is limited to users 18 and older.

For creators scoring AI video, a full-length, 44.1 kHz stereo model with a per-song API price changes the math on temp tracks and beds. At $0.08 a song, a batch of cues could undercut a stock-library subscription if the output holds up, though the API does not support the Batch, Flex, or caching discounts other Gemini models offer, and a fixed-length video may still need a re-prompt or a clip model.

Lyria 3.5 is live now in the Gemini app at gemini.google.com/music, in Flow Music, AI Studio, and Vids; API access needs a paid Gemini API key.