Rerunning LLM benchmarks

[ml-evs]

As expected, several new SoTA LLMs have been released since our preprint. For resubmission, I would like to benchmark:

• Gemini 2.5
• OpenAI gpt-4.1
• Claude Sonnet 4

I also want to try and be a bit more lenient in the benchmarking, i.e., running the prediction prompt multiple times and averaging.

[ml-evs]

c.f. https://www.nature.com/articles/s41524-025-01536-2