AI models supported by Voxt
Voxt separates speech recognition, local language models, translation models, and remote providers. This catalog describes the model families currently documented in Voxt, what each family is useful for, and where to verify the upstream project.
The short version
- Voxt separates ASR, local LLM, translation, and remote-provider channels.
- Local choices include Whisper, Qwen3 ASR, Parakeet, Nemotron, MOSS, SenseVoice, Qwen3.5, Gemma 4, GLM 4, LFM2, and Hy-MT2.
- Remote providers and OpenAI-compatible local services such as Ollama, oMLX, and LM Studio are optional configuration paths.
- The catalog reflects current public Voxt documentation; model visibility, model IDs, and hardware requirements can change with releases.
Choose a model family
Speech recognition models
Local and remote ASR choices for live dictation, recordings, meetings, and multilingual speech-to-text.
Browse this categoryLocal language models
On-device LLM choices for cleanup, formatting, rewrite, prompts, and private voice workflows.
Browse this categoryLocal translation models
Local translation choices for spoken input and selected text, including the Hy-MT2 family.
Browse this categoryRemote model providers
BYOK providers and local OpenAI-compatible services for users who want a remote or separately hosted model.
Browse this categoryHow to read this catalog
A model appearing here means it is documented as a Voxt integration or model path, not that every model is the right default for every Mac. “Local” describes where the model step can run when configured locally; it does not guarantee that every workflow, provider, or model variant is offline. Follow the upstream source link and the current in-app model picker before installing a model for production work.
FAQ
What models does Voxt support?
Voxt documents separate model channels for speech recognition, local language-model enhancement, local translation, and remote providers. The current catalog includes Whisper, Qwen3 ASR, Parakeet, Nemotron, MOSS, SenseVoice, Qwen3.5, Gemma 4, GLM 4, LFM2, Hy-MT2, and optional remote or OpenAI-compatible providers.
Can Voxt run AI models locally on a Mac?
Yes. Voxt supports local ASR, local LLM, and local translation channels when the required model and runtime are available. Local processing depends on the selected channel, model size, macOS version, and Mac hardware.
How should I choose a Voxt model?
Choose the channel first: ASR for speech recognition, LLM for cleanup or rewrite, and translation for language conversion. Then compare a few real samples using the same prompt, microphone, language pair, and hardware rather than relying on a universal model ranking.
Model setup and context