Model Traces

Model Traces provides run-level analysis of any model deployed or integrated in your account. Use it to review request inputs, generated outputs, response times, and run metadata for troubleshooting and performance investigation.

Navigation: Go to Settings → Monitoring → Model Traces.

---

Performance Metrics Summary

The metrics summary at the top of the page shows key performance data for the selected model and period.

Metric	Description
Total Requests	Total runs serviced by the model since deployment. Reflects processing speed and efficiency.
Response Time	P90 and P99 thresholds — the response times below which 90% and 99% of requests fall. Lower values indicate consistent speed; higher values indicate potential issues.
Failure Rate	Percentage of requests that failed out of total requests. For example, 5 failed out of 100 equals a 5% failure rate.
Hosting Credits	Credits consumed by the deployed model based on usage, for comparison against actual utilization.

Hosting Credits apply only to open-source and fine-tuned models. This metric is not shown for external models.

Model Traces Table

The table displays all runs for the selected model, sorted from the latest to the oldest execution date. It includes data from the first execution onward — whether from deployment (open-source and fine-tuned models) or integration (external models).

Successful requests are marked with a green check icon; failed requests with a red alert icon.

Column	Description
Status	Success or failure icon for the run.
Request ID	Unique identifier for the run record.
Response Time	Time taken by the model to respond.
Deployment Version	Model version deployed in your account.
Source Type	Type of source that initiated the request.
Source	Specific source name that sent the request.
Executed on	Run execution timestamp. Click the Sort icon to reorder records by ascending or descending date.
Input	Input text provided for the run.
Output	Output text or response generated by the model.

To view the detailed trace for a run, click its row in the table.

Export Data

Click Export to generate a CSV file of model traces records based on the selected date range and filters.

Trace Details

Select a trace record to open the Traces detail view, which displays the input, output, and metadata for that run.

• Input: The request text provided to the model using input tokens.
• Output: The text output or response generated by the model using output tokens.

Plain text is the default display format. Enable JSON to view the complete request or response payload, including additional keys not visible in plain text. Content is read-only — select the Copy icon to copy to your clipboard.

In JSON mode, the model name includes the deployment version for open-source and fine-tuned models, and the connection name for external models.

Metadata

Field	Description
Request ID	Unique identifier for the trace.
Base Model	The model used for the run.
Connection Name	The connection used to invoke the model.
Response Time	Time taken to generate the response.
Input Tokens	Number of tokens in the input.
Output Tokens	Number of tokens in the output.
Executed On	Timestamp of when the run occurred.
Source Type	The type of source that triggered the run.
Source	The specific source that triggered the run.