Hi @Kk17 , it really depends on what you value most. If you value leaderboard scores and model output quality above all else, then GPT-4 is generally the stronger of the two models. However, GPT-4 also costs nearly $40 per million tokens, your requests and responses need to be sent to a proprietary third party model, and you cannot fine tune the model on your own data.
DBRX, by contrast, is slightly lower in model performance out of the box, but can be fine tuned - either on Databricks or on another platform - to achieve performance even higher than that of GPT-4 on the tasks that you care about most. DBRX can be downloaded locally and served from any sufficiently large machine. It also costs substantially less, at around $1+ per million tokens when self served and has a context window that is 4x longer than GPT-4: nearly 33k tokens vs 8k.
