ElevenLabs just pushed Music v2.5 live across its app and API, claiming a noticeable jump in audio fidelity. The company says the new model produces tracks that sound “fuller and more natural,” a bold claim backed by internal data. In a blind test involving 47,885 comparison pairs, listeners consistently preferred v2.5 over previous iterations. The improvement wasn’t uniform across all genres, though. R&B, Soul, Hip-Hop, Rock, and orchestral compositions saw the biggest gains, suggesting the model has developed a sharper ear for complex instrumentation and vocal nuance.
Access isn’t locked behind a paywall, but the limits are strict. Free users get five lossless downloads daily, while Pro subscribers can pull up to 400 tracks a month. Crucially, users keep the rights to their generated songs. Commercial use is permitted, though free-tier creators must provide attribution. ElevenLabs has also built in guardrails: you can’t download tracks based on specific existing artists, and imitating real musicians is strictly prohibited. This approach tries to thread the needle between creative freedom and copyright compliance, a balancing act many competitors struggle to maintain.
The timing is no accident. Suno, a direct rival, released new models this week amid ongoing lawsuits over training data. ElevenLabs distinguishes itself by stating it trained existing models on “licensed stems and music,” though it hasn’t disclosed the specific sources. This contrasts sharply with Suno’s legal battles over unauthorized copyrighted content. However, don’t confuse this release with ElevenLabs’ recent licensing deal with Universal Music Group. The company clarified that the UMG partnership applies only to future, separate products, not Music v2.5. For now, v2 remains available alongside the new version, giving users a fallback if they prefer the older style. The real test will be whether these licensed datasets can scale without sacrificing the raw creativity that drives viral AI hits.
