Abstract Detecting somatic mutations from the patients’ tumor tissues has the clinical impacts in medical decision making. Library preparation methods, sequencing platforms, read alignment tools and variant calling algorithms are the major factors to influence the data analysis results. Understanding the performance of the tool combinations of the somatic variant calling pipelines has become an important issue in the use of the whole exome sequences (WES) analysis in clinical actions. In this study, we selected four state-of-the-art sequence aligners including BWA, Bowtie2, DRAGMAP, DRAGEN aligner (DragenA) and HISAT2. For the variant callers, we chose GATK Mutect2, Sentieon TNscope, DRAGEN caller (DragenC) and DeepVariant. The benchmarking tumor whole exome sequencing data released from the FDA-led Sequencing and Quality Control Phase 2 (SEQC2) consortium was applied as the true positive variants to evaluate the overall performance. Multiple combinations of the aligners and variant callers were used to assess the variation detection capability. We measured the recall, precision and F1-score for each combination in both single nucleotide variants (SNVs) and short insertions and deletions (InDels) variant detections. We also evaluated their performances in different variant allele frequencies (VAFs) and the base pair length. The results showed that the top recall, precision and F1-score in the SNVs detection were generated by the combinations of BWA+DragenC(0.9629), Bowtie2+TNscope(0.9957) and DRAGMAP+DragenC(0.9646), respectively. In the InDels detection, BWA+DragenC(0.9546), Hisat2+TNscope(0.7519) and DragenA+DragenC(0.8081) outperformed the other combinations in the recall, precision and F1-Score, respectively. In addition, we found that the variant callers could bias the variant calling results. Finally, although some combinations yielded high accuracies of variant detection, but some variants still could not be detected by these outperformed combinations. The results of this study provided the vital information that no single combination could achieve superior results in detecting all the variants of the benchmarking dataset. In conclusion, applying both merged-based and ensemble-based variants detection approaches is encouraged to further detect variants comprehensively.