Canopy Labs: Orpheus 3B

Canopy Labs: Orpheus 3B is a text-to-speech model from Canopy Labs. It is a 3B parameter model fine-tuned for English speech synthesis, with a focus on natural prosody and expressive delivery. The model was released by the provider on 23 April 2026. Its main category is speech, and it is positioned as a specialist model rather than a general purpose language model.

The model takes text as its only input and produces speech as its only output. It has a context length of 4096 tokens, and the maximum output is 3686 tokens. Orpheus 3B ships with 7 preset voices: tara, leah, jess, leo, dan, mia and zac. The provider describes it as fine-tuned for natural prosody and expressive delivery, which means the emphasis is on how the generated speech sounds rather than on text understanding or reasoning.

Those capabilities point to a fairly specific set of use cases. Narration is a natural fit, since the model is tuned for prosody and expression across longer passages of English text. Voice assistants are another, where the seven preset voices let you pick a consistent speaker for spoken responses. The same applies to any application that needs to turn written content into audio, such as reading out articles, generating spoken prompts, or producing voice tracks for a fixed set of characters or presenters. Because the input is text only, the model does not accept audio for cloning or transcription, and because the output is speech only, it does not return text you can post-process.

You call the model through api14's OpenAI-compatible API using the model id "canopylabs/orpheus-3b-0.1-ft". The request follows the same shape as other OpenAI-compatible speech endpoints, with your text as the input. The model lists no supported parameters, so you should not expect to pass sampling controls, voice selection flags or other tuning options through the request. The seven voices are part of the model's fixed configuration rather than request parameters.

The 4096 token context and 3686 token output cap are worth planning around. Very long documents will need to be split into chunks before synthesis, and the English-only focus means other languages are out of scope. If your task is text generation, reasoning, translation or anything that returns text, a general language model will fit better. Orpheus 3B is the right choice when the deliverable is spoken English audio and you want the prosody handled by the model itself.