NVIDIA introduced Nemotron 3.5 Transcribe, an open-weight speech-augmented language model (SALM) that supports 40 languages by unifying transcription, speaker attribution, translation, and contextual audio understanding in a single checkpoint.
Released on October 28, NVIDIA Nemotron 3.5 Transcribe is built as a 34B-parameter hybrid Mamba-2 Transformer mixture-of-experts (MoE) system with approximately 4.3B active parameters (including the audio encoder).
Meet NVIDIA Nemotron 3.5 Transcribe 🎙️
An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint.
Watch it keep track of who said what when everyone’s talking at once.
1/ 🧵
— NVIDIA AI (@NVIDIAAI) October 28, 2026
According to NVIDIA, it can transcribe audio in 40 languages, attribute speech to individual speakers (even in overlapping conversations), translate to or from English, and adjust its output based on natural-language instructions such as using specific terminology, producing cleaner written-style text, or redacting sensitive information.
“A single model performs verbatim, written-form, contextual, speaker-attributed, and rich transcription, speech translation, and spoken language identification in 40 languages, on recordings up to one hour long, and returns speaker-attributed word-level timestamps on request. The transcription style is chosen with a natural language instruction for each request, without retraining,” wrote NVIDIA in a blogpost. The language backbone is the Nemotron 3.5 Lightning 30B-A3B model (roughly 3.5B active), paired with a Parallel Expert Encoder.
NVIDIA released the model’s weights on Hugging Face on Wednesday claiming that it provides transcription across 40 languages with leading quality for high-resource languages (English, Spanish, French, German, Hindi, Italian, Korean, Polish, Portuguese, Russian, etc, while production-ready for others (Japanese, Thai, Turkish, Ukrainian, Bulgarian).
Also Read: NVIDIA’s $6 Billion Poolside Deal to Compete with Domestic and Chinese Frontier AI Models
