Salesforce unveiled SOV-Transcribe-1 on November 12, 2026, a speech transcription model whose weights are available on Hugging Face. It records a 2.3% word error rate on the AA-WER Index, ranking seventh among 44 evaluated models and first among open-weights models. Its closest open-weights competitors, Mistral AI’s Voxtral Small and NVIDIA’s Nemotron 3.5 Transcribe, both score 2.8%. The model has 26.6 billion parameters, of which 4.6 billion are active at any given time, and achieves a hosting endpoint speed factor approximately 60 times that of conventional systems. It forms part of Salesforce’s Sovereign AI initiative and Missionforce, which focuses on controlled AI deployment for enterprises, governments and national security workloads.
Source: Read the original article
