I've been diligently trying to use Gemini 1.5 Pro, and it is not even on the level of Llama3-70B. I really hope Gemini improves, even if it gets reduced context length.