An AI model pool groups models that meet the requirements for a task. A coding pool, for example, may contain models with suitable code quality, tool use and context limits, while a fast classification pool may use smaller models.
A router selects among eligible models in the pool using ranking, latency, cost or availability rules. Pool membership should be versioned and reevaluated because model behavior, pricing and service reliability can change.
ELI5
A model pool is a list of AI models approved for a particular kind of request. Every model in the pool should meet the task's minimum requirements for capability, policy and technical compatibility.
For example, a coding pool can contain three models that pass the same repository tests. A router can choose among them using current speed, cost and availability, while models that cannot safely use the required tools stay outside the pool.
Why create task-specific model pools?
They prevent unsuitable models from receiving a request and let the router compare only candidates that meet task requirements.
Can one model belong to several pools?
Yes. A model can be eligible for multiple task types when it satisfies the capability and operating constraints of each pool.
