The text-to-speech market has a new speed champion. Cartesia’s Sonic-3.6 has claimed the #1 position on both Artificial Analysis TTS leaderboards, marking the first time a single provider holds the top two spots simultaneously. The breakthrough positions Cartesia as the definitive alternative to ElevenLabs, the long-dominant leader in AI voice synthesis.
What makes Sonic-3.6 different
Sonic-3.6 builds on Cartesia’s state-space model architecture—a lineage originating from Albert Gu, the co-creator of the Mamba architecture. This foundation enables the model to generate speech with remarkable efficiency. The key metrics tell the story: sub-90ms time-to-first-audio means listeners hear the first sounds nearly instantly after triggering generation, compared to typical 200-300ms delays in competing systems.
Speed translates directly to cost savings at scale. Sonic-3.6 generates 136 characters per second—nearly triple ElevenLabs’ 46.7 characters per second—while charging approximately half the price. For developers building voice applications at enterprise scale, these efficiency gains compound into significant operational savings.
The leaderboard sweep
Securing #1 on both TTS leaderboards simultaneously represents an unusual achievement. Most providers compete strongly in either quality or speed, rarely both. The Artificial Analysis benchmarks evaluate TTS systems across multiple dimensions including naturalness, pronunciation accuracy, and latency. Cartesia’s dominance across both metrics suggests the state-space approach has matured into a practical advantage.
The company now holds positions 1 and 2 on the combined leaderboard, creating distance between itself and established competitors. This positioning reflects both technical progress and aggressive pricing—a combination that’s reshaping expectations in the TTS market.
Implications for the voice AI ecosystem
Sonic-3.6’s launch arrives at a pivotal moment for voice AI. Real-time voice applications—from conversational agents to accessibility tools—demand ever-lower latency. The sub-90ms threshold enables use cases previously impractical, including live translation, interactive voice response, and real-time content narration.
With support across 44 languages, Sonic-3.6 also challenges the localization market. Developers building multilingual applications previously faced a tradeoff between speed, quality, and cost. Cartesia’s pricing and language coverage eliminate at least two of those constraints.
The TTS market, long dominated by ElevenLabs, now shows clear signs of fragmentation. As competing providers close the quality gap while offering price and speed advantages, the market appears headed toward commoditization—a trajectory familiar from the LLM pricing wars of 2025-2026.