It seems like the best way of figuring out how strong a new model is, is to look at the benchmarks published by a 3rd competitor.

Want to know how well the new Google model performs compared to GPT-4? Look at the Claude benchmark table.