Skip to main content
Language models take your transcribed text and format or transform it according to your mode’s instructions. There are two types: cloud models run on hosted servers, and local models run entirely on your device. Usage of every model below is covered by your license, with no API keys needed.
Speed is a relative rating from 1 to 10. The Benchmark column is each model’s published general-capability score, not a Superwhisper measurement; see How We Benchmark for what we do measure ourselves.

​
Superwhisper (Cloud)

Hosted by Superwhisper and optimized for low latency and high-quality text processing.

​
Superwhisper (Local)

S1-mini runs entirely on your device, on macOS and Windows (Windows ARM devices need a cloud model instead). Pair it with a local voice model and nothing you dictate leaves your machine. It also works offline.

​
Anthropic (Cloud)

Anthropic’s Claude models, available through Superwhisper.

​
OpenAI (Cloud)

OpenAI’s GPT models, available through Superwhisper.

​
Groq (Cloud)

Fast inference hosted on Groq’s LPU hardware.

​
Choosing a model

S1-Language is the default and the right choice for most modes. Pick S1-mini when your mode uses a local voice model and you want the whole pipeline on-device. Reach for a larger model (GPT-5, Claude 4.5 Sonnet) only when a mode does heavy rewriting and quality matters more than speed. You can also add models from your own providers with your own API keys under Settings → Models library → Bring your own keys.