BioMysteryBench: Results as Published

BioMysteryBench tests whether a model can do real bioinformatics research work on messy real-world data. Domain experts wrote 99 questions, each built from a dataset and a controlled, objective property of that data, so every answer has an objective ground truth. Grading looks at the final answer only. The questions are split into Human Solvable (76: at least one of up to five domain experts answered it from scratch) and Human Difficult (23: no expert on the panel solved it), and publishers report the two splits separately; each problem is attempted five times and accuracy is averaged.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.