Cartesia has just released Sonic-3, their new flagship Text to Speech model for realtime conversation Sonic-3 delivers Text to Speech streaming with built-in prosodic elements such as laughter, as well as emotions like surprise and curiosity. The new model supports 42 languages, including 9 Indian languages. Sonic-3 leverages a State Space Model (SSM) architecture like Cartesia’s previous releases. Cartesia has been a leader in using SSMs, with other models typically using transformer-based architectures. SSMs are generally more compute and memory efficient, which likely contributes to Sonic-3’s comparatively impressive latency. Cartesia offers a subscription and credits pricing model ranging from $0 per month for personal use, up to $299 per month for their Scale package. Sonic-3 has been added to the Artificial Analysis Text to Speech Arena for blind preference voting. Listen to sample clips below 🎧

Introducing the AI Speech Generation Explorer — a new way to compare AI-generated speech from leading Text to Speech models The AI Speech Generation Explorer is a web app that allows users to listen to and compare AI-generated voices across leading Text to Speech systems. It’s built for developers, researchers, and creators to evaluate model capabilities and select the most suitable voices for applications such as voice agents, video narration, audiobook production, game characters, and accessibility tools. Users can explore speech outputs across four real-world categories of prompts: Customer Service, Knowledge Sharing, Digital Assistants, and Entertainment. Filters for accent, gender, and model make it easy to focus on specific voice types or compare multiple voices from the same model side by side. This launch marks another step in expanding our work within the Speech domain, where AI voice generation is advancing applications in interaction design, creative media, and inclusive technology.

Avenger 0.5 Pro ranks #2 in Image to Video in the Artificial Analysis Video Arena (All Models Leaderboard), trailing only Kling 2.5 Turbo! Avenger 0.5 Pro is the latest release from Video Rebirth, a Singapore-based video generation foundation model startup that just came out of stealth. This represents a significant upgrade from their Avenger 0.5 model released in July. While Avenger 0.5 Pro is not publicly accessible (hence why it is not on the Current Models leaderboard), Video Rebirth has indicated that their next model release is expected in the coming weeks and will be available to users. We'll provide full coverage and benchmarks once the new model launches! See how Avenger 0.5 Pro compares to other leading video models in our Video Arena 🧵