Benchmarks often test biological knowledge or narrow skills. The tasks in LifeSciBench test whether models can reason from evidence, work with scientific artifacts, handle uncertainty, and make useful decisions under real-world constraints. GPT‑Rosalind scores above GPT‑5.5
LifeSciBench tests reasoning; GPT‑Rosalind outperforms GPT‑5.5
By
–
