Primary use stated or inferred cautiously from official documentation.
N
NVIDIA / nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B
Nemotron 3.5 Transcribe 30B A4B
A speech model for transcription in 40 languages, translation and speaker diarization. It accepts audio and text and returns text. The model card reports 34B total parameters and 4.3B active including the audio encoder; the 30B name identifies its language backbone.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B |
|---|---|
| Model type | Voice and audio |
| Access model | Open weights |
| License | OpenMDW 1.1 |
| Deployment | Self-hosted |
| Release | 28 OCT 2026 |
| Knowledge cutoff | Not published |
| Entrance | Audio · Text |
| Exit | Text |
| Context window | 65,536 tokens; recordings up to one hour |
| maximum output | Not published |
| Reasoning | Configurable; disabled for transcription in the recommended protocol |
| Published tools | |
| Structured Outings | Text with speaker labels and optional word-level timestamps |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · NeMo Speech · vLLM · Transformers |
Process authorized recordings, meetings and media files, adapting vocabulary and transcription style through instructions.
This is an offline model, not a streaming ASR service. Support varies by language and task; audio understanding was evaluated only in English. Speaker labels do not identify people. Check consent, sensitive data and transcription errors.
03 / CAPABILITIES
What it can do
Orientation
Process authorized recordings, meetings and media files, adapting vocabulary and transcription style through instructions.
Context
65,536 tokens; recordings up to one hour · Not published
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
Input modalities
AudioText
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
The BF16 checkpoint declares OpenMDW 1.1. The card describes FP8 and NVFP4, but their public downloads could not be verified. No commercial per-token pricing has been verified. Budget for GPU, memory and operation.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
The evaluation separates transcription WER from speaker-attributed cpWER. Accelerated NIM performance with NVFP4 and DFlash2 does not describe the open reference server or arbitrary GPUs. October 28 is the release date; October 12 is when results were captured.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| Open ASR Leaderboard English · 9 public sets | 4,71 % | WER (%) · lower is better | View source ↗ |
| Open ASR Leaderboard multilingual · 7 languages | 4,62 % | WER (%) · lower is better | View source ↗ |
| Open ASR Leaderboard long-form | 10,98 % | WER (%) · lower is better | View source ↗ |
| AMI + CALLHOME + Mixer 6 + NOTSOFAR-1 · full sessions | 15,80 % | cpWER (%) · lower is better | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Nemotron 3.5 Transcribe 30B A4B
Version added to Inferama's verified catalog.
07 / SOURCES