Providers and Models
OSTT can use local Parakeet and Whisper models or cloud providers. Run ostt model to choose the active model. Run ostt auth login first when you want to use a cloud provider.
Transcription request params are configured under [provider.params] or [provider."model".params], or passed per invocation with --param key=value. Use ostt model params PROVIDER/MODEL to list supported params for a local or cloud model. Params under [provider.params] apply to every model of that provider, so each one must be supported by the model you select; put model-specific params under [provider."model".params].
For practical recommendations by use case, see Choosing a Transcription Model and Cloud Models.
Local
Local models run on your machine and do not require an API key. Open the local model manager with:
ostt modelChoose Local models to download curated models, activate downloaded models, delete model files, inspect metadata, or add a custom Hugging Face/direct model URL.
The curated model list is maintained in the kristoferlund/ostt-models repository. Open a pull request there to suggest a model for inclusion in the default list shown to all users.
Local transcription has full GPU support through Metal on macOS and CUDA or Vulkan builds on Linux.
| Provider | Models | Params |
|---|---|---|
parakeet | parakeet/parakeet-ultra, parakeet/parakeet-tdt-0.6b-v3, parakeet/parakeet-tdt-0.6b-v2, parakeet/parakeet-tdt_ctc-110m | None; the language is detected automatically |
whisper | whisper/turbo, whisper/large-v3, regional KB-Whisper and NB-Whisper models, and more | language, no_context, temperature, no_speech_thold, prompt |
Local models need 16 kHz mono 16-bit WAV. OSTT records in that format automatically when a local model is active; files you pass to ostt transcribe must already be in it. See Local Models.
Local Whisper params use the same provider/model format as cloud providers, for example [whisper.params], [whisper.turbo.params], or --param language=sv. Use model IDs such as whisper/turbo or parakeet/parakeet-ultra. See Local Models for setup details and local param descriptions.
Cloud Provider Guides
| Provider | Models | Guide |
|---|---|---|
| OpenAI | openai/gpt-transcribe, openai/gpt-4o-transcribe, openai/gpt-4o-mini-transcribe, openai/gpt-4o-transcribe-diarize, openai/whisper-1 | OpenAI |
| Deepgram | deepgram/nova-3, deepgram/nova-3-medical, deepgram/nova-3-pharma, deepgram/nova-2 | Deepgram |
| Groq | groq/whisper-large-v3, groq/whisper-large-v3-turbo | Groq |
| DeepInfra | Whisper Large V3, Whisper Large V3 Turbo, Qwen3 ASR, Nemotron, and Voxtral speech-recognition models | DeepInfra |
| AssemblyAI | assemblyai/universal-3-5-pro, assemblyai/universal-2 | AssemblyAI |
| Berget | berget/klang/pianissimo, berget/KBLab/kb-whisper-large, berget/NbAiLab/nb-whisper-large, berget/openai/whisper-large-v3 | Berget |
| ElevenLabs | elevenlabs/scribe_v2, elevenlabs/scribe_v2_medical, elevenlabs/scribe_v1 | ElevenLabs |
| Mistral | mistral/voxtral-mini-latest, mistral/voxtral-mini-2602 | Mistral |
To run another engine yourself, such as faster-whisper or an OpenAI-compatible server, use the command and http providers described in External Engines.
Landing Pages
These pages are broader introductions and workflow overviews: