autosearch:model-routing

LLM model-tier routing assigns tasks to optimal model tiers, such as Fast, Standard, or Best, to balance cost and quality during runtime decision-making across processing pipelines.

You assign model tiers by providing each leaf skill with a frontmatter model_tier annotation, which steers the runtime routing without triggering any automatic model switching.

Yes, you can escalate or de-escalate LLM tasks at runtime based on task complexity, latency constraints, and cost considerations using safe escalation policies that preserve runtime prompts.

The best way to optimize LLM costs is applying tiered routing to ensure the most economical model is used for routine steps while reserving higher tiers for critical synthesis tasks.

No, model-tier routing does not automatically switch LLMs during execution; it relies on frontmatter model_tier guidance to steer runtime decisions and prevent overcommitment of resources.