Researchers published DeepSomatic, a deep‑learning platform that improves somatic small‑variant detection across short‑ and long‑read sequencing technologies, and released a set of public benchmarking datasets (CASTLE‑panel) to support method development. The work appeared in Nature Biotechnology and addresses a persistent gap in accurate somatic calling across platform types. The authors trained and validated DeepSomatic on a variety of sequencing inputs and demonstrated performance gains in calling SNVs and indels in cancer samples. They also produced seven reference benchmarks from cancer cell lines and made them publicly available to help the field standardize evaluation and accelerate tool adoption. Accurate somatic variant calling is essential for tumor profiling, clinical decision‑making, and trial enrollment; the paired release of a caller and community benchmarks should reduce assay discordance and improve diagnostic reproducibility across labs using different sequencing platforms.