Hey!
It depends a lot on your application. Different LLMs excel in different functions. Proprietary models will be more expensive then Open Source ones, but they will likely lead you to a faster time to production with your applications.
Also, if you are thinking about RAG or Fine Tuning a model, this also influences the model you choose. There are lighter models that takes less time to fine tune, and models that will work better with your latency requirements when working with RAGs.
I recommend checking some benchmarks - in this case, you'll see benchmarks for maths, language understanding, etc. I also recommend checking out Hugging face's model page, where you can pick models based on their objective task and popularity.
Finally, you can also leverage Databricks Playground to compare multiple models and how they respond to the same prompts.
