Artificial intelligence model ranking assigns an order or score to models that are eligible for a request. The ranking can be fixed by policy, calculated from evaluation results or adjusted dynamically using current speed, health and cost data.
A router may try the highest-ranked available model or select the fastest candidate that meets a quality threshold. Ranking criteria should remain transparent enough to explain a selection and should be checked for stale or biased evidence.
Acronyms and aliases
ranked model selection synonymAI model ranking variantartificial intelligence model ranking variant
Related terms
Frequently asked questions
How are AI models ranked for routing?
They can be ranked by task quality, reliability, cost, latency, user preference or a weighted combination of those signals.
Is model ranking the same for every request?
Not necessarily. A dynamic router can change the ranking according to the task, user, service health and time budget.
