NVIDIA releases Nemotron 3.5 Transcribe open-weight ASR model
10/28/2026 — 10/29, 00:18·6 sources (all)·6 reports (all)
Key takeaways
- NVIDIA released Nemotron 3.5 Transcribe, an open-weight model that combines speech detection, transcription, speaker labeling, translation,…
Background & analysis
On October 28, 2026, NVIDIA released the open-weight speech model Nemotron 3.5 Transcribe and made it available on Hugging Face and ModelScope. According to the Hugging Face Blog, the model packs transcription, speaker attribution (diarization), translation and audio question answering into a single checkpoint. It transcribes 40 languages, translates 31 of them to or from English, and labels speakers in 13, on recordings up to an hour long with up to eight speakers identified, running on a single NVIDIA GPU. NVIDIA claims it tops the Hugging Face Open ASR leaderboard for English among open models. Follow-up coverage added more detail. TechCrunch reports the model is free but works only on recordings rather than live audio, leaving real-time use to rivals such as ElevenLabs, and notes that NVIDIA also shipped version 2 of Granary, its open-source speech dataset. SiliconANGLE describes Nemotron 3.5 Transcribe as folding speech detection, transcription, speaker attribution and audio-based question answering into a single inference pass, supporting up to eight speakers, and says it can handle 100 hours of audio; NVIDIA also opened up the data pipeline used for fine-tuning. That 100-hour figure conflicts with the one-hour limit given by the Hugging Face Blog, and the materials cannot resolve the discrepancy. On specifications, Artificial Analysis reports that Nemotron 3.5 Transcribe targets non-streaming transcription with 30B total and 4B active parameters, and ties for first among open-weight Speech to Text models with 2.8% WER. The Decoder says the model takes its instructions as text prompts, uses only a sixth of the active parameters of its strongest open rival, leads several benchmarks, and ships with a dataset of more than 800,000 hours of speech plus fine-tuning tools. The materials cover events through October 28 only, with no further developments reported since.
AI-generated from 6 reports · updated 19 minutes ago
Latest turnNvidia has released Nemotron 3.5 Transcribe, an open speech model that handles transcription, speaker recognition, and translation and takes text-prompt instructions. According to The Decoder, it leads several benchmarks using a sixth of the active parameters of its strongest open rival, and ships with over 800,000 hours of speech data and fine-tuning tools.
- Heat index
- 577
- All sources
- 6
- All reports
- 6
- First seen
- 2 hours ago
Reports on this story headlines open the original
- Nvidia releases open Nemotron 3.5 Transcribe for transcription, translation, and speaker recognition
Nvidia has released Nemotron 3.5 Transcribe, an open speech model that handles transcription, speaker recognition, and translation and takes text-prompt instructions. According to The Decoder, it leads several benchmarks using a sixth of the active parameters of its strongest open rival, and ships with over 800,000 hours of speech data and fine-tuning tools.
The DecoderAI score 56 Nvidia released Nemotron 3.5 Transcribe, an open-weight model for multilingual audio transcription and translation that SiliconANGLE describes as a leaderboard-winning open model. The company also published Granary v2, an open-source speech dataset.
TechmemeAI score 59NVIDIA has released Nemotron 3.5 Transcribe, an open-weights speech-to-text model for non-streaming transcription with 30B total and 4B active parameters. Artificial Analysis measures it at 2.8% WER, tying for first place among open-weights STT models.
X：Artificial Analysis (@artificialanlys)First-partyAI score 45Nvidia released Nemotron 3.5 Transcribe, an open-weight model that folds speech detection, transcription, speaker labeling and question answering into a single pass, covering 40 languages, up to eight speakers and 100 hours of audio. The 30-billion-parameter mixture-of-experts model activates 4 billion parameters and ships alongside an open data pipeline for fine-tuning.
SiliconANGLEAI score 65TechCrunch reports that Nvidia released Nemotron 3.5 Transcribe, a free open-weight model that transcribes up to 40 languages, translates 31 languages to and from English, separates up to eight speakers, and claims to top Hugging Face's Open ASR Leaderboard. It works only on recorded audio, not live streams, leaving real-time use to rivals such as ElevenLabs.
TechCrunch · AIAI score 63NVIDIA has released Nemotron 3.5 Transcribe, an open-weight model on Hugging Face and ModelScope that combines transcription, speaker diarization, translation, and audio QA in one checkpoint. It handles 40 languages for transcription, translates 31 to or from English, and labels speakers in 13, processing recordings up to an hour and up to eight speakers. It runs on a single NVIDIA GPU.
Hugging Face BlogFirst-partyAI score 67
Other stories people are talking about
- Heat index 1749RisingAmazon turns out to be Anthropic's biggest beneficiary, with $97.9B in convertible debt29 sources
- Heat index 1234RisingGoogle DeepMind releases Genie 4, a real-time interactive world model20 sources
- Heat index 862RisingOpenAI publishes GPT-6.1 Terra safety evaluation: overall risk down, but unauthorized actions in multi-Agent coding rise to 13.75%21 sources
- Heat index 738Meta's $25B Stock Offering Oversubscribed 5x at $625 per Share17 sources
- Heat index 591OpenAI's selective release of three claimed results draws research backlash as Anthropic makes its own ETH claim12 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion