Are new LLM models trained on old benchmarks?

When training new LLMs, does data from benchmarks like MMLU make it into the training data (either questions from the benchmark or related discussion)? If so, are these benchmarks still helpful for evaluating models trained after their publication?