Deepgram: Nova-3

Deepgram Nova-3 is a speech-to-text model from Deepgram, listed on api14 under the model id deepgram/nova-3. It sits in the transcription category, and Deepgram describes it as a general-purpose speech-to-text model with both monolingual and multilingual transcription support. If your application needs to turn recorded or streamed speech into written text, this is the kind of model the provider built it for.

The model takes audio as its input and returns a transcription as its output. That is the whole shape of the interface: you send audio, you get text back. The provider description calls out monolingual and multilingual transcription, so it can handle speech in a single language as well as material that spans more than one. There is no context window figure listed for this model, and the maximum output is listed as zero, so plan around the transcription result itself rather than around a chat-style token budget.

Because the only input is audio and the only output is a transcription, the practical uses follow directly. You can use Nova-3 to produce transcripts of meetings, interviews, voice notes, podcasts or call recordings. Multilingual support makes it a fit for teams whose audio is not all in one language, for example customer calls or field recordings that mix languages. The transcripts it returns can then feed whatever comes next in your pipeline, such as search indexing, summarisation by a separate text model, or storage for review.

You call the model through api14's OpenAI-compatible API using the model id deepgram/nova-3. Because the endpoint follows the OpenAI format, you can point an existing OpenAI client at api14 and send your audio for transcription without rewriting your integration. The model lists no supported parameters, so do not assume tuning options such as temperature or custom prompt fields are available; send the audio and rely on the defaults the gateway applies. Check the live price table on the page for current rates before you commit to a workload.

One thing to weigh is scope. Nova-3 is a transcription model, not a general text or reasoning model, so if your task is generating, editing or analysing text rather than converting speech to text, a text model will fit better. If your audio is in a language or mix of languages outside what the provider supports, verify results on a sample before rolling it out. For straightforward speech-to-text work, the model id deepgram/nova-3 is the one to use.