Yeah it’s a shame a lot of these benchmarks are behind. My favourite was ‘SlopCodeBench’ [1] as I’m most interested in ai reinforcing its own bad decisions, but it’s not even up to current gen oai

1: https://www.scbench.ai/