Search
On February 5, 2026, Mistral AI, a French artificial intelligence firm, introduced the Voxtral Transcribe 2 series models. This series comprises two distinct versions: Voxtral Realtime and Voxtral Mini Transcribe V2. Voxtral Realtime utilizes a streaming architecture, achieving an incredibly low latency of under 200 milliseconds. Moreover, it offers support for 13 languages. In a move that promotes open research and development, the model weights are made publicly available under the Apache 2.0 license. On the other hand, Voxtral Mini Transcribe V2 is tailored for batch processing tasks. It has the capability to process audio recordings of up to 3 hours in length in a single go. Notably, its accuracy outperforms that of models such as GPT-4o mini Transcribe. When it comes to pricing, its API is offered at a rate ranging from $0.003 to $0.006 per minute.
-
C114 Communication Network -
Communication Home
