After the release of GPT4o mini I decided to migrate some of my implementaton to this new model. Main reason was these implementations where very token consuming ( RAG pipelines).
The results were dreadful.
- Very hard to have the model follow instructions
- Results of function call were very unpredictable
- The model introduced some markdown (I think) that I was not aware of
- The reasoning of the model was very bad
The pro’s
- THe model responded very fast (via Azure OpenAI that it)
- It is cheap. Very cheap
I decided to go back to GPT4o after one day
What are your experiences with this model. What are valid usecases?