ns-ui / router-tier-cascade
Support agent — model routing, last 15 minutes
Every request enters at the cheapest tier and runs right until it hits that tier's notch. If it is harder than the notch allows it spills one row down and keeps going. Where a request stops is what it cost.
tierdifficulty spill →accepted · spend
edge-2b$0.0004
0$0
edge-2b: ceiling 22 percent, 0 requests accepted, $0 spent.fast-8b$0.0025
0$0
fast-8b: ceiling 46 percent, 0 requests accepted, $0 spent.pro-70b$0.0150
0$0
pro-70b: ceiling 72 percent, 0 requests accepted, $0 spent.frontier$0.0750
0$0
frontier: ceiling 93 percent, 0 requests accepted, $0 spent.each row spills at its notch — median per request on hover
drag a notch (or Tab to it and use ← →) to re-route settled traffic