Routing & Cost
Govern model choices and prove the savings.
Requested vs selected model, estimated vs actual cost, and router-only vs full-gateway latency - shown side by side so nothing is conflated.
Arena Router Mode
Model selection only - no provider round-trip, no audit. Leaderboard / embed path.
p50
-
p95
-
p99
-
No samples yet - run traffic or the benchmark suite to populate.
Gateway Mode
Full governed request: auth, policy, shield, proxy, audit, proof, provider round-trip.
p50
-
p95
-
p99
-
No samples yet - run traffic or the benchmark suite to populate.
Latency is process-local and measured. Arena and Gateway numbers are never combined - Arena excludes the provider call by design.
Decisions
0
Est. savings
$0
Actual cost
$0
Avg decision latency
-
Model fan-out
Candidate models allowed by policy - cost, quality, and latency tier. The router's most-used pick is highlighted.
Router cost over time
Requested cost vs selected/actual cost from routing decisions.
Routing decisions
Requested cost is estimated; actual cost is measured from provider usage when reported.
Loading routing decisions…
Estimated savings (per decision): … Cost-proof carries four explicit labels: requested cost, estimated savings, actual provider-billed cost, and actual savings.