"Better" is a moving target since LLMs tend to be compared on large suites of tasks and datasets, and over time the consensus of effective benchmarks changes.

But the question will resolve positively if it shows matching or better performance on at least 3 benchmarks that are reasonably standard at the time and somewhat general purpose (not specialized on maths, coding or some other narrow class of reasoning).

Open-source means that both the code and the weights are openly available, regardless of the license.