Voxtral transcribes at the speed of sound.
Get smart on it
Mistral AI released two new speech-to-text models called Voxtral Transcribe 2, which converts spoken audio into written text with high accuracy and minimal delay. One model, Voxtral Realtime, is designed for live applications with configurable latency down to under 200 milliseconds, while the other, Voxtral Mini Transcribe V2, handles batch processing of recorded audio up to 3 hours long. Both models support 13 languages and include features like speaker identification, word-level timestamps, and context biasing for technical terms. The models are positioned as more cost-effective and accurate alternatives to competing transcription services, with Voxtral Realtime released as open-weights software under the Apache 2.0 license.
Related news
Vibe coding platform Base44 launches own model as AI startups seek defensibility
Wix-owned vibe coding platform Base44 has started rolling out its own AI model — with hopes that it will eventually outperform frontier models.
DiScoFormer: One transformer for density and score, across distributions
DiScoFormer is a transformer model that estimates both the density and score of a data distribution in a single forward pass without requiring retraining for new distributions. The density describes where data points cluster, while the score, the gradient of log-density, points toward more probable regions and is used in diffusion-based generative models, Bayesian sampling, and scientific simulations. Existing methods force a trade-off: classical kernel density estimation works on any distribution but loses accuracy in high dimensions, while neural score-matching models stay accurate in high dimensions but must be retrained for each new distribution. DiScoFormer significantly outperforms kernel density estimation, cutting score error by 6.5 times and density error by more than 37 times in 100 dimensions, while generalizing to distributions with shapes and complexities not seen during training.
