ElevenLabs launches Scribe v3 for context-aware speech transcription
ElevenLabs released Scribe v3 and Scribe v3 Live, new speech recognition models that handle background noise and crosstalk accurately. The models use conversational context to prevent voice agents from cutting off callers mid-sentence. Benchmarks show they beat OpenAI on accuracy but are slower and more expensive than competitors like Microsoft and xAI.
Summarized by AI from one outlet
Source
- THE DECODERRead original ↗
Latest
- Models
Sakana AI Begins In-House Model Development
Sakana AI has announced it is starting to develop its own AI models from scratch, a shift from its previous focus on utilizing existing models. This move marks a new strategic direction toward fundamental model research and development.
- Models
Liquid AI releases LiquidCodec streaming audio codec
Liquid AI released LiquidCodec, a 74M-parameter streaming audio codec powering LFM3-Audio-2B. It runs at 12.5 Hz and encodes 361x faster than real-time on a single CPU thread, enhancing multimodal model performance.
- Models
OpenAI Teases New Cybersecurity AI
OpenAI previewed a new cybersecurity AI model during its 'Frontier at the Cyber' livestream. The model is reportedly twice as performant as Mythos 5.2 and is expected to demonstrate enhanced defense and detection capabilities.
- Models
NVIDIA releases synthetic datasets for Japanese AI
At the 'NVIDIA Agentic AI Day Japan 2026' event in Tokyo, NVIDIA introduced two synthetic datasets, 'Nemotron-Personas-Finance-Japan' and 'Nemotron-PII-Japan', designed to support the development of Japanese AI models and accelerate training.
- Models
LLMON: New LLM-Native Markup Language for Granite 4.1 Models
Hugging Face has introduced LLMON, a markup language designed specifically for large language models. It is now supported in Granite 4.1 3B and 8B models, aiming to improve how LLMs handle structured data.
All news → · All rights belong to the original creators