Voxtral Transcribe 2 consists of two speech-to-text models with transcription quality, diarization, and ultra-low latency.

Google’s ATLAS study reveals how languages help each other in AI training, offering scaling laws and pairing insights for better multilingual models.