Muse offers access to models with different capabilities in different languages to best fit your needs. Increasingly powerful models are slower and more expensive, but they can tackle more tasks more effectively.
You can select which model to use by setting the
X-Model header in your API call. Models codenames are formatted as
orion-fr is the Orion model for French for instance).
Models can be further specialized by using 🤹 Skills.
Model classes & capabilities
We offer three classes of models, with increasing capabilities. All models support all API endpoints, however skills are models specific.
|Class||Description||Trained on...||Best for...|
|Our smallest and simplest models. Available across many languages, suitable for simple straightforward tasks.||filtered web||Simple classification with |
| ||A great tradeoff between power and cost. Can be specialized with skills to perform specific tasks at ||curated data||Most text generation & classification use cases.|
| ||Our most powerful and capable models.||curated data||Complex tasks, specialized text generation.|
We recommend starting your experiments with the most powerful model available, and then optimize for the smallest size that still fulfills your needs. If a skill matching your task is available, it can be used to allow smaller models to perform on-par with larger ones.
Languages & model availability
✅ are available in the API, 📆 are coming soon, and ❌ are not available.
|French [||English [||Spanish [||Italian [||German [|
Don't see a language or model you need? Shoot us an e-mail at firstname.lastname@example.org.
🟢 are recommended models, 🔵 are new models, 🟡 are models in end-of-life, 🔴 are retired models, and 📆 are coming soon.
|Model||Training data||Data cut-off||Status|
We recommend you make the switch from 🟡 end-of-life models to 🔵 new models as soon as possible.