SOV-Transcribe Models by Salesforce AI Research
Salesforce AI Research releases · 1 in the last 12 monthsChangelog →
1 model2026
Last refreshed 2026-11-13. Next refresh: weekly.
Details
ResearcherSalesforce AI Research
LicenseApache 2.0OSI-approved
Commercial usePermitted
Models1
Released2026
Capabilities
MultimodalAll models
About
SOV-Transcribe is Salesforce's family of open-weight speech recognition models from its Sovereign AI team. The first release, SOV-Transcribe-1 (November 12, 2026), is a speech-augmented language model built on a Whisper large-v3 encoder and Gemma 4 26B A4B, released under Apache 2.0 for self-hosted transcription.
Decision facts
- Best fit
- audiospeech recognitionvision and multimodal work
- Capability starting point
- SOV-Transcribe-1 with multimodal inputs
- Lowest tracked input
- Not tracked
- Closest related family
- AssemblyAI
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
1 in view
SOV-Transcribe-1Current
Use when the workload needs speech recognition, multimodal inputs, and audio.
2026-11speech recognitionmultimodal inputsaudio
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| SOV-Transcribe-1 | Use when the workload needs speech recognition, multimodal inputs, and audio. | 2026-11 | speech recognitionmultimodal inputsaudio | Current |
Release Timeline
1 release group2026-11
1 current
SOV-Transcribe-1
Currentspeech recognitionmultimodal inputsaudio
Specifications(1 models)
| Model | Released | Parameters | Multimodal |
|---|---|---|---|
| SOV-Transcribe-1 | 2026-11 | 26.6B (4.6B active) | Yes |