Hacker News Reader

benchmarks dont show the quality or 'correctness' of the response though.

What do they show?

If you mean this particular benchmark, it shows how much people like the responses a LLM gives.