{"id":"nvidia-nemotron-3-5-transcribe-30b-a4b","slug":"nvidia-nemotron-3-5-transcribe-30b-a4b","name":"NVIDIA Nemotron 3.5 Transcribe","orgId":"nvidia","family":"nvidia-nemotron-3-5-transcribe-30b-a4b","releaseDate":null,"availability":"ga","openWeights":true,"license":"OpenMDW-1.1","paramsTotal":30000000000,"paramsSource":"slug","contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B","hfRepo":"nvidia/Nemotron-3.5-Transcribe-30B-A4B","modalities":["audio","speech"],"aliases":["NVIDIA Nemotron 3.5 Transcribe","nvidia/nvidia-nemotron-3.5-transcribe-30b-a4b","nvidia-nemotron-3.5-transcribe-30b-a4b","nemotron-3.5-transcribe-30b-a4b","nemotron-3.5-transcribe","nemotron 3.5 transcribe","Nvidia Nemotron 3.5 Transcribe 30B A4b","nvidia nemotron 3.5 transcribe"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-10-28","announced":"2026-10-28","blurb":"NVIDIA's open speech model that puts transcription in 40 languages, speaker attribution, translation, text-prompted context and audio question answering in one checkpoint. It is a mixture-of-experts model built on the Nemotron 3.5 Lightning 30B-A3B language model with an ASR encoder and a speaker encoder, and handles recordings up to an hour long, offline (not streaming). Weights on Hugging Face under OpenMDW-1.1, which allows commercial use.","api":"none","openWeights":"yes","params":{"text":"34B total, about 4.3B active, including the audio encoder (the 30B-A4B in the name refers to the language backbone)","source":"NVIDIA's Hugging Face model card"},"facts":[{"label":"Task","value":"Speech-to-text with speaker attribution, word timestamps, speech translation, contextual prompting and audio understanding"},{"label":"Languages","value":"Transcription in 40 languages; translation for 31 to or from English; speaker attribution in 13, up to 8 speakers"},{"label":"Architecture","value":"Mamba-2/Transformer hybrid MoE language backbone (Nemotron 3.5 Lightning 30B-A3B) with a parallel expert encoder: a 600M Transformer ASR encoder plus a ~99M Nemotron-3-Diarization speaker encoder, and a CTC timestamp head"},{"label":"Input length","value":"65,536-token context; recordings up to 1 hour; offline only"},{"label":"Weights","value":"Hugging Face nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B (BF16), with FP8 and NVFP4 variants and a DFlash2 draft model; release date 10/28/2026 per the card, also on NGC and in NeMo Speech"},{"label":"License","value":"OpenMDW-1.1; commercial and non-commercial use"},{"label":"Hardware","value":"Runs on one H100 80 GB, per the card"},{"label":"API","value":"The card lists a build.nvidia.com release; NVIDIA's Hugging Face blog names hosted endpoints from Baseten, DeepInfra and Simplismart"},{"label":"Released with","value":"Granary v2 speech dataset"},{"label":"Technical report","value":"NVIDIA Nemotron 3.5 Transcribe Technical Report (research.nvidia.com)"}],"claims":["Lab claim: first among open-weight models on the Hugging Face Open ASR Leaderboard for English, with 4.71% mean WER vs 4.94% for Qwen3-ASR-1.7B.","Lab claim: 10.98% long-form WER vs 13.37% for Parakeet-TDT-0.6B-v3, 4.62% vs 6.49% on a seven-language multilingual set, and 15.80% cpWER on multi-speaker audio (NVIDIA's own evaluation setup)."],"sources":[{"kind":"repo","title":"nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B on Hugging Face","url":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B","date":"2026-10-28"},{"kind":"lab","title":"NVIDIA on Hugging Face blog: Nemotron 3.5 Transcribe and Granary v2","url":"https://huggingface.co/blog/nvidia/nemotron-transcribe-granary","date":"2026-10-28"},{"kind":"paper","title":"NVIDIA Nemotron 3.5 Transcribe Technical Report","url":"https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3.5-Transcribe-Technical-Report.pdf"},{"kind":"lab","title":"NVIDIA AI on X: Nemotron 3.5 Transcribe","url":"https://x.com/NVIDIAAI/status/2115462300241141882","date":"2026-10-28"}],"firstBenchmark":{"date":"2026-10-29","benchmarkId":"open-asr"}},"eligibleSince":"2026-10-29","organization":{"id":"nvidia","name":"NVIDIA","slug":"nvidia","country":null,"url":null,"color":"#76B900","mark":"N"},"price":null,"answer":{"word":"#8","scope":"Best current rank: #8 in Speech-to-text, as of Nov 3, 2026.","lead":"As of Nov 3, 2026, NVIDIA Nemotron 3.5 Transcribe is ranked #8 in Speech-to-text (since Nov 3, 2026; sources Open ASR Leaderboard (English short-form); methodology v1.0). Ranked #8 in Speech-to-text. That board does not award a crown.","confidence":"low"},"status":"BENCHMARKED","verdicts":[{"id":"v_overall_stt_nvidia-nemotron-3-5-transcribe-30b-a4b_2026-11-03_open_RANKED","model":"nvidia-nemotron-3-5-transcribe-30b-a4b","model_name":"NVIDIA Nemotron 3.5 Transcribe","category":"stt","category_name":"Speech-to-text","slice":"overall","state":"RANKED","rank":8,"present_evidence":[],"missing_evidence":[],"coverage":null,"segments":[],"confirm":[],"margin":null,"confidence":"low","valid_from":"2026-11-03","valid_to":null,"configuration":null,"rationale":"Ranked #8 in Speech-to-text. That board does not award a crown.","safeguard_flags":[],"co_sota_with":[],"challengers":[],"methodology":"1.0","label":"#8 in Speech-to-text"},{"id":"v_open-weights_stt_nvidia-nemotron-3-5-transcribe-30b-a4b_2026-11-03_open_RANKED","model":"nvidia-nemotron-3-5-transcribe-30b-a4b","model_name":"NVIDIA Nemotron 3.5 Transcribe","category":"stt","category_name":"Speech-to-text","slice":"open-weights","state":"RANKED","rank":1,"present_evidence":[],"missing_evidence":[],"coverage":null,"segments":[],"confirm":[],"margin":null,"confidence":"low","valid_from":"2026-11-03","valid_to":null,"configuration":null,"rationale":"Ranked #1 in Speech-to-text. That board does not award a crown.","safeguard_flags":[],"co_sota_with":[],"challengers":[],"methodology":"1.0","label":"#1 in Speech-to-text (open weights)"},{"id":"v_le-32b_stt_nvidia-nemotron-3-5-transcribe-30b-a4b_2026-11-03_open_RANKED","model":"nvidia-nemotron-3-5-transcribe-30b-a4b","model_name":"NVIDIA Nemotron 3.5 Transcribe","category":"stt","category_name":"Speech-to-text","slice":"le-32b","state":"RANKED","rank":1,"present_evidence":[],"missing_evidence":[],"coverage":null,"segments":[],"confirm":[],"margin":null,"confidence":"low","valid_from":"2026-11-03","valid_to":null,"configuration":null,"rationale":"Ranked #1 in Speech-to-text. That board does not award a crown.","safeguard_flags":[],"co_sota_with":[],"challengers":[],"methodology":"1.0","label":"#1 in Speech-to-text (≤32B)"}]}