Two companion Nature Biotechnology reports introduce DeepSomatic, a deep‑learning somatic small‑variant caller designed to work across short‑ and long‑read sequencing platforms, and a set of public benchmarking datasets (CASTLE‑panel). The tools address persistent gaps in consistent somatic variant detection by supplying a cross‑platform caller and standardized benchmark data derived from cancer cell lines. DeepSomatic improves sensitivity and specificity for detecting somatic SNVs and indels across sequencing technologies. The CASTLE benchmarks provide seven validated somatic datasets to evaluate pipelines. Together, they lower a key barrier in cancer genomics: reproducible identification of tumor mutations from heterogeneous sequencing inputs. The work will influence clinical labs, variant‑calling tool development, and regulatory validation of genomic assays.