NVIDIA: Nemotron 3.5 ASR Streaming Multilingual 0.6B

NVIDIA: Nemotron 3.5 ASR Streaming Multilingual 0.6B is a speech recognition model from NVIDIA. It belongs to the Nemotron 3.5 family and is distributed through api14 under the model id nvidia/nemotron-3.5-asr-streaming-multilingual-0.6b. Its design centres on automatic speech recognition rather than general text generation, so it is best understood as a dedicated transcription component rather than a chat model.

The model takes audio as its input and produces a transcription as its output. Its architecture is a prompt-conditioned, cache-aware FastConformer-RNNT design, which is built for streaming recognition with low latency. It is multilingual, targeting more than 40 languages, and the streaming behaviour means audio can be transcribed incrementally as it arrives rather than only after a complete recording is available. No context length in tokens is published for this model, and the maximum output is listed as zero, which reflects that its output is a transcription stream rather than a token budget in the usual sense.

Those characteristics point to a fairly specific set of uses. Real-time captioning is the clearest fit, since the streaming design is intended to emit text while speech is still in progress. The same applies to live transcription for meetings, broadcasts or any audio feed where waiting for the full file would be too slow. Because the model covers more than 40 languages, it also suits applications that serve multilingual audiences and need a single recognition component rather than one model per language. Voice interfaces that need to convert spoken input into text before passing it to a downstream system are another natural use.

You call the model through api14's OpenAI-compatible API using the model id nvidia/nemotron-3.5-asr-streaming-multilingual-0.6b. Because the interface follows the OpenAI format, existing clients and SDKs that already target that shape can be pointed at api14 without a rewrite. The model accepts audio input and returns a transcription. No request parameters are listed as supported for this model, so you should not assume options such as temperature, top_p or similar sampling controls are available; send the audio and read the transcription back.

It is worth checking that a dedicated ASR model is what your application needs. This model does not take text input and does not produce general text completions, so tasks like summarising a transcript, answering questions about it or translating it require a separate language model downstream. If your audio is not in one of the supported languages, or if you need a text-in, text-out model, a different model on api14 will fit better.