Turns out that many current science-based LLM benchmarks have flaws in their answers. When corrected, the LLM benchmark scores rose significantly.

Many current science-based LLM benchmarks have been found to contain flaws in their answers. When these benchmarks were corrected, the scores of the LLMs being evaluated rose significantly. This discovery suggests that the true capabilities of LLMs in scientific domains might be underestimated due to issues within the benchmarks themselves. This information comes from a paper posted on Arxiv, as discussed in a dev_community on reddit.com.