Select a Model Routing Strategy for Mixed-Complexity Tasks

ROLE Act as an AI systems procurement and routing advisor. OBJECTIVE Design a model-selection strategy for [TASK PORTFOLIO]. CONTEXT Example inputs and desired outputs: [TASK EXAMPLES]. Quality requirements: [QUALITY REQUIREMENTS]. Latency and budget constraints: [RESOURCE LIMITS]. Data restrictions: [DATA RESTRICTIONS]. INSTRUCTIONS Group tasks by reasoning difficulty, context length, modality, structured-output needs, and consequence of error. Identify tasks that should use deterministic rules instead of a language model. Propose routing signals available before execution, plus escalation conditions based on independently checkable output defects. Do not use a model's self-reported confidence as the sole gate. Compare candidate model classes and describe how to evaluate named options using representative samples. Account for retries, escalation, context caching where supported, and human review in the cost model. OUTPUT FORMAT Deliver a routing table, candidate evaluation rubric, illustrative cost formula, and a time-boxed comparison experiment. QUALITY CHECK Flag assumptions, avoid fabricated statistics, and label unverified model capabilities. State when routing complexity would cost more to maintain than a single-model approach.