Nemotron 3.5 Transcribe ties for top open-weight STT at 4B active params
- Source
- x.com
- Date
NVIDIA's Nemotron 3.5 Transcribe ties for #1 among open weights Speech to Text models with 2.8% WER, using just 4B active parameters Nemotron 3.5 Transcribe is NVIDIA's new open weights Speech to Text model for non-streaming transcription, with 30B total and 4B active parameters, and the most accurate NVIDIA model on our non-streaming Speech to Text leaderboard. Key takeaways ➤ Accuracy: Nemotron 3.5 Transcribe scores 2.8% on the AA-WER Index, tied with Mistral's Voxtral Small for #1 among open-weight models. It leads Voxtral Small on AA-AgentTalk, our dataset of voice agent conversations, at 1.8% vs. 2.3%. ➤ Model size: With 4B active parameters (30B total), Nemotron 3.5 Transcribe sits on the open weights frontier of AA-WER vs. active parameters. It matches the accuracy of Mistral's Voxtral Small with 6x fewer active parameters (4B vs. 24B), while other open weights models of a similar size sit at around 4% WER. Congratulations to @NVIDIAAI on the release! See more details below ⬇️
Context
Nemotron 3.5 Transcribe, a new speech-to-text model from NVIDIA, matches the most accurate open-weight transcription model in Artificial Analysis's testing while running far fewer parameters. In a post on the result, Artificial Analysis, which runs model leaderboards, reports that the model scores 2.8% on its AA-WER Index, tied for first among open-weight models with Mistral AI's Voxtral Small. WER, or word error rate, measures how many words a transcript gets wrong relative to what was actually said, so lower is better. On AA-AgentTalk, Artificial Analysis's dataset of voice conversations, Nemotron 3.5 Transcribe leads Voxtral Small at 1.8% against 2.3%.
The size comparison is what sets the result apart. Nemotron 3.5 Transcribe has 30B parameters in total but 4B active parameters, the part of the network that actually does the computing on a given input. Artificial Analysis says it matches Voxtral Small's accuracy with 6x fewer active parameters (4B against 24B), while other open-weight models of a similar size sit at around 4% WER. Artificial Analysis places it on the open-weights frontier when accuracy is plotted against active parameters. The model is built for non- transcription, meaning it works on complete recordings rather than live audio, and Artificial Analysis says it is the most accurate NVIDIA model on its non-streaming leaderboard.
These are Artificial Analysis's measurements on its own benchmarks. Only the opening post of the thread was collected; the follow-up posts it points to were not, so this account covers accuracy and model size only. NVIDIA's own write-up of Nemotron 3.5 Transcribe is covered separately.
- open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
articleNVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapthuggingface.co
postMistral's Voxtral 3 speech models benchmarked: top-3 STT, mid-pack TTSArtificial Analysis
postNVIDIA has just released the first Nemotron 3.5 model: Nemotron 3.5 Lightning, aArtificial Analysis
Checking sign-in…
Loading comments…