A city's 311 department is evaluating several Amazon Bedrock foundation models for a resident-service chatbot that must handle long conversation histories and respond fluently in the languages spoken by the city's resident population. What should guide the department's model selection?

Selecting a foundation model against stated requirements means comparing published capability specs — context-window length and supported languages — to the department's actual needs, since Bedrock offers multiple models that differ meaningfully on both dimensions. A model released more recently isn't guaranteed to have the longest context window or the strongest multilingual coverage; recency is not a capability spec, so defaulting to 'newest' skips the actual evaluation. Choosing on price alone ignores that a cheaper model failing to meet language or context requirements produces a chatbot that mishandles resident conversations, which costs more in rework and complaints than the token savings are worth. Parameter count is a rough proxy for general capability, but it doesn't reliably predict context-window length or multilingual performance — those are architecture and training-data decisions independent of raw size. Scope caveat: published specs describe general capability, not guaranteed performance on this department's specific dialects or jargon, so specs should be paired with a pilot test. Operational check: run a sample of representative multilingual, long-history conversations against shortlisted models and measure whether responses stay accurate as the conversation grows.