NVIDIA launches Nemotron 3.5 Transcribe with a free browser demo

NVIDIA announced Nemotron 3.5 Transcribe on October 28, 2026. The open-weight speech model takes recorded audio and written instructions. NVIDIA says it transcribes 40 languages, translates 31 to or from English, and labels speakers in 13. It supports recordings up to one hour and up to eight speakers.

You can try the official Hugging Face browser demo. In our recorded test on October 29, no account or credits were needed. This is a hosted demo, not a promise of unlimited free service or local processing in your browser.

A written prompt can change how the model formats a transcript or asks for speaker labels. That makes it worth testing on interviews, meetings and recorded conversations before building a workflow around it.

We uploaded a five-minute excerpt from the 1984 Reagan-Mondale debate. The demo captured Reagan's youth-and-inexperience punchline. Our short shows the real output and plays the matching source audio.

The speaker labels changed from SPK:1 on Reagan's punchline to SPK:0 on his continuation. The demo also rendered panelist Trewhitt's name as 'Truett'. These are findings from our recorded run, not a measure of the model's overall accuracy.

The demo does not offer a native speaker-renaming control. Its model-generated numbered labels are static. The Reagan and Still Reagan callouts and magnifier in our short are FreeStack annotations, not native tool labels.

Sources: NVIDIA announcement and NVIDIA's Hugging Face release post, checked October 29, 2026.

Debate audio: Reagan Library, 'Second Presidential Debate with President Reagan and Walter Mondale, October 21, 1984'. Excerpts trimmed, sped up and volume adjusted. Original source via Wikimedia Commons. License: CC BY 3.0.