🏠

Gradium AI: The Voice Revolution 🤯🎙️

September 01, 2026 | Author

AI

🎧 Audio Summaries
English flag
French flag
German flag
Japanese flag
Korean flag
Mandarin flag
Spanish flag
🛒 Shop on Amazon

🧠Quick Intel


  • Gradium AI launched a new text-to-speech model on August 31, 2026, making it the default across its API and Studio.
  • The new Gradium model achieved an 81.0% human-rated pass rate on a 500-sentence hard-case set across five languages, surpassing Cartesia Sonic 3.6 (75.1%) and ElevenLabs v3 Conversational (65.4%).
  • Time to first audio is 216 ms at P50 on Coval, 170 ms faster than the previous model.
  • Gradium open-sourced a 500-sentence evaluation set on Hugging Face under CC BY 4.0, comprising 100 items across 10 criteria in five languages (EN, DE, FR, ES, PT).
  • The Gradium model’s P50 time to first audio is 216 ms, while Inworld TTS 2 posts a 166 ms median.
  • Existing Gradium users do not require any action; they can reuse existing voice IDs by installing the Python SDK and pointing it at the WebSocket TTS endpoint.
  • Gradium is offering 1M credits for complete hard-case failure reports submitted on its Discord.

    • 📝Summary


    On August 31, 2026, Gradium AI announced the release of a new text-to-speech model, integrating it as the default across its API and Studio. The company’s evaluation, using a 500-sentence set across five languages, achieved an 81.0% human-rated pass rate, surpassing Cartesia Sonic 3.6 and ElevenLabs v3 Conversational. Time to first audio reached 216 ms on Coval, 170 ms faster than the previous model. Gradium released a publicly available evaluation set, fostering transparency. Existing users require no action, while new teams can utilize the Python SDK. The company is actively seeking feedback through a Discord channel, offering 1M credits for detailed failure reports.

    💡Insights

    ▼


    GRADIUM TTS: A NEW STANDARD IN TEXT-TO-SPEECH ACCURACY
    Gradium AI has recently launched a significant advancement in text-to-speech technology with the release of their new voice agent model, marking a pivotal shift in the industry. This model has demonstrated exceptional performance, achieving an impressive 81.0% human-rated pass rate on a challenging “hard-case” set of 500 sentences across five languages – a considerable leap ahead of competitors like Cartesia Sonic 3.6 (75.1%) and ElevenLabs v3 Conversational (65.4%). Crucially, this transition occurred seamlessly on August 31, 2026, without any required migration for existing users, ensuring uninterrupted service. The core innovation lies in a meticulously crafted evaluation set, openly available on Hugging Face under a CC BY 4.0 license, allowing for rigorous testing and validation.

    TECHNICAL PERFORMANCE AND BENCHMARKING
    The Gradium TTS model’s performance isn’t just about accuracy; it’s also about speed and efficiency. Notably, the time to first audio is a remarkable 216 ms at P50 on the Coval platform, 170 ms faster than the model it replaced. This speed advantage is further reinforced by a tight interquartile range of 30 ms across 480 runs, showcasing a consistent and reliable performance. Comparative testing against other leading models – Cartesia Sonic 3.6 (454 ms median with 165 ms spread), Inworld TTS 2 (166 ms median), Fish Audio S2.1 Pro (291 ms), and ElevenLabs v3 Conversational (329 ms) – clearly positions Gradium TTS as a leader in both speed and accuracy, particularly when considering the “hard-case” scenarios. The focus on minimizing variance – the 30 ms p75-p25 range – underscores Gradium’s commitment to delivering a consistently high-quality experience.

    IMPLEMENTATION AND COMMUNITY ENGAGEMENT
    The deployment of the new Gradium TTS model is designed for simplicity and ease of integration. Existing users simply need to install the Python SDK, point it at the WebSocket TTS endpoint, and continue utilizing their existing voice IDs. Gradium is actively fostering a community around this technology, offering 1 million credits for detailed reports of hard-case failures via their Discord channel. Furthermore, Gradium encourages engagement through various channels: a detailed release post, access to the comprehensive dataset, and active participation on Twitter and their 150k+ member ML SubReddit. They also maintain a newsletter for subscribers seeking updates and insights.

    Related Articles