Google's late night expansion strategy, throwing out three major AI heavyweight updates at once! From February 18th to 20th, Google DeepMind officially launched the music generation model Lyria3, which was synchronized with Google Music to complete the function launch. The flagship model Gemini3.1 also received an epic update, covering music creation, multimodal reasoning, and programming agent scenarios, directly hardening GPT-5.2 and Claude 4.6, reconstructing the competitive landscape of the AI industry.
Lyria3 explosion! Integrate Gemini to generate professional grade music in 30 seconds
As the latest flagship music generation model of Google DeepMind, Lyria3 has not been launched separately, but is directly integrated into the entire Gemini platform without additional downloads or queuing tests. Hundreds of millions of users can trigger music generation directly through Gemini dialog boxes, completely reducing the threshold for AI music creation.
Core Capability: Supports multimodal input of text, images, and videos. Upload a sunset photo, a pet video, or input a style description to generate 48kHz stereo audio within 30 seconds. With one click matching of vocals, instruments, and lyrics, emotions are coherent and rhythmic, and the sound quality is comparable to professional recording standards;
Creative lyric generation: Adopting a guided lyric generation mode, users do not need to input lyrics word by word. Simply describe the theme and tone, and Gemini will automatically write lyrics that fit the melody. It can also incorporate specified phrases to balance rhythm fluency and creative needs. The only regret is that it currently does not support word for word replication of lyrics and specific artist voice imitation to avoid copyright risks;
Invisible watermark protection: Each generated audio segment is embedded with a SynthID watermark that is imperceptible to the human ear, even after MP3 compression and tone shift, it will not disappear. It can be identified by AI generated identity through dedicated tools, solving the problem of copyright tracing.
Google Music linkage! Seamless connection, free unlocking of short video music
Along with Lyria3, there is also a new feature upgrade from Google Music, which deeply links the two to form a complete closed loop of "AI generation music application ecological distribution", accurately adapting to short video creation scenarios.
After the launch of Google Music, users can directly import the audio generated by Lyria3 into the platform with one click for secondary editing such as editing, mixing, and adding sound effects. It supports downloading MP3 pure audio or MP4 with cover versions, perfectly adapting to the 30 second music requirements of short video platforms such as YouTube Shorts and TikTok - it should be noted that the 30 second duration is not a technical limitation, but a product strategy of Google's targeted layout of the short video track, accurately hitting the high-frequency pain points of creators.
In addition, Google Music has also partnered with YouTube's Dream Track tool to expand it globally, allowing creators to directly generate exclusive AI music for shorts without additional licensing, significantly reducing the cost of music usage for short video creation and impacting the market share of traditional material music libraries.
Gemini3.1 Slaughterhead Update! Doubling reasoning, crushing GPT-5.2/Claude 4.6
After the launch of Lyria3 and Google Music, Gemini3.1 immediately welcomed an epic update. As the strongest flagship model of Google at present, its performance has doubled, and multiple benchmark scores have topped the global authoritative charts, surpassing similar competitors in strength.
Running score champion: In the rigorous ARC-AGI-2 test, Gemini3.1 achieved an incredible high score of 77.1%, more than twice the performance of the previous generation 3.0 Pro, far surpassing Claude Opus 4.6 (68.8%) and GPT-5.2 (34.5%); In the AAII comprehensive evaluation, the total score leads Claude Opus by 4.6 points, but the API call cost is less than half;
Comprehensive capability leap: Supports 1 million Token ultra long context, which is currently one of the few models that can complete the ultimate test of 1M Token. GPT-5.2 and Claude 4.6 do not support it; The gap between programming and agent capabilities is leading, and it has taken the first place in tests such as LiveCodeBench Pro and Terminal Punch 2.0, with a significant decrease in illusion rate and more accurate logical reasoning;
Deep integration at the bottom layer: Lyria3's music generation and Veo video generation capabilities are integrated at the bottom layer, allowing users to complete the entire process of "text → music → video" creation in the Gemini dialog box without switching tools, achieving seamless collaboration of multimodal capabilities.
Competitive showdown! Lyria3 vs Suno/Udio， Advantages and weaknesses can be seen clearly at a glance
The most anticipated of Google's three major updates this time is undoubtedly the direct competition between Lyria3 and mainstream AI music models such as Suno and Udio. Each of the three has its own focus and accurately covers different user needs, without absolute crushing, but can stand out with its ecological advantages.
In terms of shortcomings, the 30 second audio output limit of Lyria3 is indeed difficult to meet the needs of complete song creation compared to the 4-minute full track supported by Suno and Udio; And currently, there is a lack of fine editing function, which makes it difficult to accurately modify specific voice parts and mixing ratios after generation, and more inclined towards lightweight scenes that are "generated and used".
But the advantages are also obvious: Lyria3 relies on Gemini's language ability, and the narrative and humorous nature of its lyrics far exceeds that of Suno and Udio. It can truly understand meter and rhythm, and the generated lyrics are more in line with the music itself; And there is no need to register a separate account or pay separately, integrated into Gemini, with extremely low distribution threshold, making it convenient for high-frequency Gemini users; The 48kHz stereo output has also reached the top level in the industry, and the sound quality is not inferior to Udio.
