Implementation of Smaller Models in the API (10-20B)

I would like to propose the addition of smaller models, ranging from approximately 10 to 20 billion parameters (or even 30), to the API.*

These models are crucial for evaluations in agentic workflows, as they offer a balance between cost-efficiency and performance that is not present in the smaller 3 billion parameter models.

Specifically, models such as Ministral 8B and Llama 8B have demonstrated fine capabilities irrespective of their sizes, making them ideal choices for various test evaluations.

By providing a range of smaller models with varying parameter counts, developers will have more options to choose from when selecting the appropriate model for their specific needs. This would enable them to fine-tune their applications more effectively, resulting in better overall performance and user experiences.