A new third-party study finds Gemini’s Pro model achieved comparable but slightly inferior accuracy compared to the current version of OpenAI’s GPT 3.5 Turbo. However, It outperforms Mixtral on every task.

Furthermore, Gemini performed better than GPT 3.5 Turbo on particularly long and complex reasoning tasks and was also adept multilingually in tasks where responses were not filtered.

source: https://arxiv.org/abs/2312.11444

However, the result on Mixteral should be taken with a grab of salt, as a user on X raised issues on the Mixteral experimental setup (https://x.com/fluffykittnmeow/status/1737044933339472254?s=20)