Selecting appropriate architectural patterns and their implementing technologies is a complex design task. We evaluate six LLMs (five open-source, one closed-source) across four realistic scenarios in three setups: (i) naive versus prompt-engineered pattern suggestion, (ii) decision-tree-guided selection using our Comprehensive Architecture Pattern Integration (CAPI) method, and (iii) mapping patterns to concrete tools drawn from a provided list with enforced JSON output. We assess answer reasonableness, consistency, architectural-pattern specificity, and output structure; with CAPI we additionally quantify coverage and distance to a desired pattern set. Findings: With only scenario descriptions, most models provide reasonable starting points; prompt engineering primarily improves focus on architectural (rather than design) patterns and consistency. CAPI guidance expands coverage and can approach human performance, but models exhibit a pronounced bias toward microservices and may over-suggest patterns, yielding higher distance scores in several scenarios. Given a curated tool list, all models propose plausible technologies, though category cues in the list influence the selected patterns. Only two models consistently produced strict, parser-ready JSON; others required light post-processing. Overall, LLMs—when paired with structured prompts and decision-tree guidance—usefully augment architectural decision-making, while highlighting needs for tighter output control and broader, less biased pattern coverage.