In n8n, a single user prompt can result in multiple model requests depending on your workflow structure. This often happens when your input is split into many chunks like in summarization chains where each chunk triggers one API call, multiple items or loops reach an AI Agent or chat model node causing it to run several times, or several tools or branches feed into the same AI node increasing how often it’s triggered.
To confirm this, test with a simple workflow using a manual trigger plus a single LLM node with no Agent, tools, or vector store. If that works reliably, the issue is likely your workflow architecture rather than the model provider. Reducing the number of tools or branches feeding the AI node, avoiding unintended loops, and cleaning up inputs so only one item reaches the AI node at a time usually helps.