Mimi Codec Interactive Demo Explained

Frisson Labs launches interactive browser demo of Kyutai's Mimi codec, enabling users to encode audio into 32 token streams and explore how semantic and acoustic information separates across codebook levels.

Moonshot AI Releases Kimi-K2.6 Model

Chinese AI startup Moonshot AI launched Kimi-K2.6, an open-source LLM with 1T parameters featuring 384 experts and MLA attention optimization, outperforming GPT-5.4 and Claude Opus 4.6 on benchmarks.

Cohere Launches Open-Source AI Transcription Model

Enterprise AI company Cohere released an open-source speech-to-text model called Cohere Transcribe, designed for real-time audio transcription in meetings and note-taking, currently ranking top on Hugging Face's speech recognition leaderboard.