In this video, I’m showing you the fastest way to pick the best LLM for your n8n workflows without burning hours testing 10+ models every time a new release drops. The n8n team shipped an official benchmarks page (and an API) that ranks 60+ models based on what actually matters inside n8n: tool use, hallucinations, logic, classification, structured output, speed, and cost. I walk through how to describe your use case (like OCR + PDF extraction) and instantly get model recommendations, plus how to copy model names for OpenRouter or grab the exact model ID. I also demo the “paste workflow JSON” approach, including the big caveat: there’s a 65,000 character limit, so you often have to paste only a section (like your AI Agent node) instead of the full workflow. Then we go hands-on with the benchmarks API inside n8n using an HTTP Request node and the “import cURL” shortcut. I cover top-model queries, category lists, single-model lookups, rate limits (5/min, 15/hour), and a current bug in the recommendation endpoint (the include_results boolean) with the simple workaround: delete that parameter. The takeaway: use benchmarks as a strong starting point, then validate in your real workflow—no hype, just builds that work.