Cohere Transcribe

The Cohere Transcribe profile supports CohereLabs/cohere-transcribe-03-2026 through vLLM v0.28.0. It provides completed microphone transcription, checkpoints, and audio-file results, including streamed file responses.

Connect the model

Model access comes first

Accept the model’s Hugging Face access conditions and authenticate the inference runtime with an account that has access. This download login is separate from Freehand’s API key for the running server.

Prepare the inference machine using the vLLM backend guide, including its audio dependencies, then serve the model:

Serve Cohere Transcribe

vllm serve CohereLabs/cohere-transcribe-03-2026

1. Add a vLLM connection using the running server’s API address.
2. Open the Settings cog in Voice transcription or Audio file and select the served model under Transcription. A custom server alias is fine.
3. Choose Cohere Transcribe as the model profile, select your language, and save.

Language and recognition controls

Control	Supported behavior
Spoken language	English, French, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Chinese, Japanese, Korean, Vietnamese, or Arabic
Server default (English)	Omits the language field; does not request automatic detection
Temperature	Optional override
Punctuation	Enabled by vLLM; no Freehand switch
Context hints and shared vocabulary	Unavailable in this vLLM integration
Realtime microphone streaming	Unavailable; recording or checkpoint requests must complete first

Settings for the direct Transformers API do not necessarily apply to this server. Cleanup and delivery use your existing workflow settings.

Set your spoken languageChoose the language used for recognition and cleanup.

Transcribe an audio fileUpload a file and review its result before copying.