It's blind test, and the code to compute scores is open. So unless they are deliberately modifying the data it should be a fair ranking. Just I think openassist

It's blind test, and the code to compute scores is open. So unless they are deliberately modifying the data it should be a fair ranking. Just I think openassistant llama(30B) version is better than vicuna 13b, but it is not included in the test.