AI news videos

NVIDIA Launches Nemotron 3.5 Transcribe, One Open Speech Model For 40 Languages

NVIDIA says it introduced NVIDIA Nemotron 3.5 Transcribe on October 28, 2026, an open weight speech model that handles transcription, speaker attribution and translation in one promptable model, and can also answer questions about a recording.

Watch on YouTube

Transcript

Sources

Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:

• PER-SENTENCE AUDIT, IN SCRIPT ORDER. (1) Hook sentence one, 'NVIDIA introduced a new open speech model today': article says 'Introduced today, NVIDIA Nemotron 3.5 Transcribe is a new, open weight speech model'. (2) Hook sentence two, 'RTX Spark laptops now have a price': article says 'RTX Spark laptops are available now from ASUS, Dell, HP, Lenovo, Microsoft and MSI, starting at $2,599'; a price now appears, where our 2026-10-07 Short noted none had been published. (3) Scene 1 sentence one, 'Nemotron 3.5 Transcribe is on Hugging Face now': article says 'It's available today on Hugging Face in the BF16, FP8 and NVFP4 formats'. (4) Scene 1 sentence two, 'One promptable model does transcription, speaker labels, translation, and questions about the audio, in forty languages': article says 'Instead of needing multiple models for transcription, speaker labels and translation, developers can use one promptable model that handles all three, plus questions about the audio. It transcribes 40 languages, including seven Indic languages.' We narrated the 40-language figure and dropped the seven Indic languages detail for length only. (5) Scene 2, 'NVIDIA says it ranks first on Hugging Face Open ASR, with about eighty percent fewer active parameters than the top open weight model': article says 'Nemotron 3.5 Transcribe ranks No. 1 on the Hugging Face Open ASR and No. 2 among open weight models on Artificial Analysis, delivering leading accuracy with approximately 4.3 billion active parameters, 80% fewer than the No. 1 open weight model'. Attribution to NVIDIA is stated in narration. (6) Scene 3 sentence one, 'A new NVIDIA App control splits GPU resources between your foreground app and a background agent': article says 'With the new resource management feature in NVIDIA App, users can now choose how GPU resources are shared. They can prioritize their foreground app for the best performance, or give their background-running agent more resources'. (7) Scene 3 sentence two, 'Laptops start at two thousand five hundred ninety nine dollars': article says 'starting at $2,599'. Spoken once in the whole script. (8) Scene 4, 'NVIDIA lists several of these updates as coming soon, and the rankings are its own, with no independent test': article's summary list is headed 'New and expected RTX Spark updates, coming soon, include' and then lists Nemotron 3.5 Transcribe, LFM3, the NVIDIA App resource controls, Phantom Blade Zero and RTX Remix, while the body separately calls Nemotron 3.5 Transcribe 'available today'; that tension is reported, not resolved. No third-party test or independent benchmark run appears anywhere in the post. (9) Endcard: Inboxsmith promotional line plus the fixed closing line; no factual claim. NVIDIA SELF-REPORTED FIGURES: the No. 1 Hugging Face Open ASR ranking, the No. 2 Artificial Analysis open-weight ranking, the approximately 4.3 billion active parameter count, the 80% parameter reduction versus the No. 1 open weight model, the 40-language and seven-Indic-language coverage, the $2,599 starting price, and the availability of 128GB configurations. All are NVIDIA's own account of its own product and of third-party leaderboards; we did not open those leaderboards. DELIBERATELY NOT NARRATED: the October 2026 NVIDIA and Microsoft partnership, the San Francisco event, preorders, the October 16 arrival date, November compact desktops, Microsoft Execution Containers, Surface Laptop Ultra and the previewed deskside AI supercomputer, because our 2026-10-07 Short already covered those; the BF16, FP8 and NVFP4 format list, NeMo-Speech.cpp, vLLM, llama.cpp, Jetson AGX Thor and the GeForce 5090 GPU name, because glued brand-plus-digit and dotted tokens are mispronounced by the voice engine; Liquid AI's new open model family and its 1.7-billion versus 2.6-billion parameter comparison, Phantom Blade Zero's launch with ray tracing and DLSS 4.5, and RTX Remix version 1.6 bringing path tracing to Minecraft Java Edition, all cut for length in favour of the lead story; the up to 1 petaflop FP4 performance, 14mm thickness and three-pound weight specifications, also cut for length. VERIFICATION STATUS: nothing in this script was verified outside this single nvidia.com blog page. There is no second source and no video transcript. The claims are accurately transcribed from NVIDIA's post; they are not independently confirmed, and the script says so on camera in scene 4.(Single first-party source: the NVIDIA corporate blog post 'NVIDIA RTX Spark Windows Laptops Gain New Models, Agent Tools and Games', by Shreevar Rastogi, published October 28, 2026, captured 2026-10-28T15:18Z. This is NVIDIA writing about NVIDIA products, so every number in it is vendor self-reporting. No video transcript was available for this topic: YouTube transcript research returned zero usable transcripts, so there is no video corroboration and no second written source. Nothing in this script was checked against any page other than this one nvidia.com blog post.)