In metagenome analysis, computational methods for assembly, taxonomic profilingand binning are key components facilitating downstream biological datainterpretation. However, a lack of consensus about benchmarking datasets andevaluation metrics complicates proper performance assessment. The CriticalAssessment of Metagenome Interpretation (CAMI) challenge has engaged the globaldeveloper community to benchmark their programs on datasets of unprecedentedcomplexity and realism. Benchmark metagenomes were generated from newlysequenced ~700 microorganisms and ~600 novel viruses and plasmids, includinggenomes with varying degrees of relatedness to each other and to publicly availableones and representing common experimental setups. Across all datasets, assemblyand genome binning programs performed well for species represented by individualgenomes, while performance was substantially affected by the presence of relatedstrains. Taxonomic profiling and binning programs were proficient at high taxonomicranks, with a notable performance decrease below the family level. Parametersettings substantially impacted performances, underscoring the importance ofprogram reproducibility. While highlighting current challenges in computationalmetagenomics, the CAMI results provide a roadmap for software selection to answerspecific research questions.