Models & Routing Eligibility
See which models CostPilot may use, what they cost, and exactly which model the existing router would select for each tier and department.
Catalog health
Warnings identify configuration that can change which model the router selects. They do not change routing themselves.
Routing preview
Uses CostPilot’s current selection priority and cascade policy. No AI call, spend, or audit event is created.
Routing outcomes
What CostPilot actually used. Older calls without exact model telemetry are explicitly marked as tier-inferred.
Routing calibration
Is the complexity classifier (keywords + token threshold) still tuned correctly? Built from every governed request's own routing decision — no outcome data, just whether the signals themselves still make sense.
Request size by tier
Scout and Analyst should rarely exceed the token threshold under normal routing — a high "over threshold" share means overrides (throttling, tags, agent bounds) are deciding more requests than the classifier itself.
| Tier | Requests | Median tokens | P90 tokens | Max tokens | Over threshold |
|---|---|---|---|---|---|
| Loading… | |||||
Keyword calibration
How often each configured keyword actually fires, and the average size of the requests it fires on. A keyword that never fires is dead weight; one firing constantly on very short requests is likely over-triggering.
| Keyword | Configured | Triggers | % of requests | Avg request size |
|---|---|---|---|---|
| Loading… | ||||
| Model | Provider | Tier | API Model ID | Input / 1M tokens | Output / 1M tokens | Status | Routing Role | |
|---|---|---|---|---|---|---|---|---|
| Loading models... | ||||||||