Cost & Latency Dashboards
ANTS Platform ships two prebuilt dashboards for spend and speed. They need no setup: as soon as traces arrive with model names and usage, the widgets fill in.
Open them from Gauge AI → Cost Dashboard and Gauge AI → Latency Dashboard. They need Gauge AI read access (or the Cost & Latency dashboards sub-module): Owner, Department Admin, Security & Compliance and Cost Analyst have it. See Members & Roles.
Cost Dashboard
| Widget | What it shows |
|---|---|
| Total costs | Spend over time across all generations. |
| Total Count Traces and Total Count Observations | Volume for the selected range. |
| Cost by Model Name | Spend per model name as reported in your traces. |
| Top 20 Users by Cost | Spend per userId on the trace. |
| Top 20 Use Cases (Trace) by Cost and (Observation) by Cost | Spend per trace name and per observation name. |
| Cost by Environment | Share of spend per environment. |
| P 95 Cost per Trace | 95th percentile trace cost over time, per trace name. |
| P 95 Input Cost per Observation and P 95 Output Cost per Observation | How expensive a single call gets on the input and output side. |
| Cost by Source | Spend over time split by whether traffic came through the Budget AI gateway. SDK and OpenTelemetry traffic has an empty source. |
| Cost by Gateway Key | Spend per Budget AI key. |
Latency Dashboard
| Widget | What it shows |
|---|---|
| P 95 Latency by Use Case | 95th percentile trace latency over time, per trace name. |
| P 95 Latency by Model | 95th percentile generation latency over time, per model. |
| P 95 Time To First Token by Model | How long streaming responses take to start, per model. |
| Avg Output Tokens Per Second by Model | Generation speed per model. |
| P 95 Latency by Level (Observations) | Latency split by observation level, for example ERROR versus DEFAULT. |
| Max Latency by User Id (Traces) | The slowest trace each user saw. |
| Avg Time To First Token by Prompt Name (Observations) | Time to first token per managed prompt. |
| P 95 Latency by Source | Latency of Budget AI gateway traffic compared with SDK and OpenTelemetry traffic (empty source). |
Filters
Both dashboards default to the last 7 days. The date range you can pick is limited by your plan's data access period. Use the filter builder to narrow by Environment, Trace Name, Observation Name, Score Name, Tags, Source, Gateway key, Gateway model group, User, Session, Metadata, Release or Version.
Customizing
The prebuilt dashboards are read-only. Roles with Gauge AI write access (Owner, Department Admin, Cost Analyst) see Clone, which creates an editable copy where you can add, remove and resize widgets. See Custom Dashboards.
Where the numbers come from
Cost is calculated per generation from usage and your model prices, or taken from the cost your traces report. Latency is the duration of each trace or observation. If a widget is empty or a cost looks wrong, check that generations carry a model name and usage, and that the model has a price definition.
Next steps
- Cost & Performance Metrics: cost and latency next to quality scores.
- AI Usage: volumes of traces, observations and scores.
- AI Optimizer: find a cheaper or faster model for a costly agent.
- Budget AI: spend limits and gateway spend analytics.