I've tried it with the 70B Llama 3 model. The results were decent nothing that impressive. I would say something between GPT-3.5 and GPT-4. I've a draft for that, but I'm looking for a good usecase to write a blog about.