NVIDIA Nemotron 3.5 Transcribe NIM Overview
Description
NVIDIA Nemotron 3.5 Transcribe NIM provides GPU-accelerated inference for Nemotron-3.5-Transcribe-30B-A4B, a speech-augmented language model developed by NVIDIA. The model combines speech and speaker-aware audio perception with an autoregressive mixture-of-experts language model to produce text from audio and task instructions.
The NIM is designed for offline multilingual transcription, speech translation, contextual and speaker-attributed transcription, and audio understanding. Applications include meeting and media processing, subtitles, localization, and speech-enabled interfaces. It processes complete recordings; it is not a realtime streaming ASR service. Available tasks, languages, optional word timestamps, and deployment limits depend on the selected model artifact and serving configuration. See the model card and deployment documentation for release-specific details.
The container components are ready for commercial or non-commercial use.
License/Terms of Use
Governing Terms: The NIM container is governed by the NVIDIA Software License Agreement and the Product-Specific Terms for NVIDIA AI Products; and use of the model is governed by the OpenMDW License Agreement v1.1.
You are responsible for ensuring that your use of NVIDIA provided models complies with all applicable laws.
Deployment Geography
Global
Release Date
- build.nvidia.com 10/28/2026: Nemotron 3.5 Transcribe
- Hugging Face 10/28/2026: Nemotron-3.5-Transcribe-30B-A4B
- NGC 10/28/2026: Nemotron 3.5 Transcribe NIM container
Program Classes
The container provides the serving software for the following NVIDIA model. Model artifacts are downloaded automatically during initialization when not already available in the configured model cache. Download-disabled deployments must provide the required artifacts in advance.
| Model Name & Link | Use Case | How to Pull the Model |
|---|---|---|
| NVIDIA Nemotron-3.5-Transcribe-30B-A4B | Offline multilingual transcription, speech translation, contextual and speaker-attributed transcription, and audio understanding. | Automatic through the NIM model-artifact initialization workflow |
Deployment Details
Follow the ASR model deployment guide for environment requirements, container startup, model-artifact access, and deployment configuration. The NIM is designed for Linux systems with supported NVIDIA GPUs, NVIDIA drivers, and NVIDIA Container Toolkit.
Select a supported precision and concurrency profile for the available GPU memory and workload. Hardware, precision, parallelism, and serving-mode support must be checked against the selected release. Enabling multi-turn serving disables MTP and DFlash2 speculative decoding.
Audio and text instructions are submitted through the documented HTTP APIs. The service returns text and, when requested and supported, structured speech metadata. Word timestamps and speaker labels are estimates; speaker labels do not establish a person's identity.
The model is designed and optimized for NVIDIA GPU-accelerated systems using CUDA libraries and GPU inference kernels. Evaluate accuracy, latency, memory usage, and throughput on representative data for your application.
References
- Nemotron-3.5-Transcribe-30B-A4B model card
- Nemotron 3.5 Transcribe on build.nvidia.com
- Nemotron 3.5 Transcribe NIM on NGC
- ASR model deployment documentation
Ethical Considerations
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal developer team
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal developer team to ensure these software components meet requirements for the relevant industry and use case and address unforeseen product misuse.
Transcripts and answers can contain omissions, hallucinations, incorrect language choices, or speaker-attribution errors. Review outputs before using them in consequential workflows. Audio, prompts, and outputs can contain personal or sensitive information; follow applicable privacy and data-handling requirements.
Please report quality, risk, security vulnerabilities, or NVIDIA AI concerns through the NVIDIA reporting program.
Get Help
Getting Started with the NIM
Visit the ASR model deployment guide for prerequisites, deployment instructions, and configuration details, and the NGC container page for release information.