We just integrated Gemini 1.5 (flash and pro) into Discourse AI.
I am very pleasantly surprised at the quality of both models, here is an example interaction:
What I liked about this is how well it followed instruction, I used to get terrible refusals in 1.0 model and it is much more eager now to follow.
I love that there are system instructions now, image support looks great as well. As it stands the Audio / Video mean Gemini models are the only ones on the market (at the moment) with support for these modalities direct from the API.