Two Nature Biotechnology reports detail major advances in somatic‑variant detection: DeepSomatic, a deep‑learning somatic small‑variant caller that works across short‑ and long‑read platforms, and CASTLE, a suite of public somatic benchmarking datasets. Together the work addresses key reproducibility and sensitivity gaps for cancer genomics by delivering methods and standards that labs can adopt for variant calling across technologies. Somatic variant calling identifies mutations acquired by tumor cells (not inherited germline variants); accurate detection is essential for precision oncology decisions. The DeepSomatic models and CASTLE benchmarks make long‑read and hybrid sequencing more actionable by improving detection of complex structural variants and indels that matter clinically.