Cost & Performance Metrics

These analytics views sit between the headline dashboards and the optimizer: they line up cost, latency, and quality so you can decide where tuning or a model swap is worth it.

Cost Performance Metrics

A detailed look at where cost meets quality:

• Per-model cost table — cost, usage count, and quality scores side by side, so you can spot models that cost a lot for little quality gain.
• Top agents/traces by volume — what's driving usage.
• Cost by user — per-user spend.
• Model usage trends — usage per model over time.
• Scores table — evaluation scores with trends.

Latency & Analytics

Performance with a quality lens:

• Latency tables — p50 / p95 / p99 by model and user.
• Generation latency over time.
• Score analytics — quality distributions alongside latency, to expose speed-vs-quality trade-offs.

Reading the trade-off

The point of these views is comparison, not a single number:

• A model that's cheap but low-quality shows up as low cost + low scores — a candidate to replace.
• A model that's expensive but no better shows high cost + similar scores — a candidate to downgrade via the AI Optimizer.
• A model that's fast but worse appears as low latency + low scores — weigh the trade.

Next steps

• AI Optimizer — get a concrete model-swap recommendation.
• Scores — how the quality numbers are produced.