Execute benchmark tests

Execute the benchmark test to measure the AI performance on predefined test cases. Test execution typically assesses retrieval accuracy, such as whether the correct document appears in the top results, along with answer quality, ensuring responses are correct, grounded in the knowledge base, and free from hallucinations. It also measures latency, including retrieval and response generation time, and checks stability to detect regressions after model, prompt, or index changes.

Procedure
  1. Navigate to Benchmarking > Tests.
    The Tests page appears.
  2. Select New.

    The Create benchmark test dialog box is displayed.

  3. Configure the benchmark test Execution options.

    Option Description

    Run now

    Select this option to run the test immediately. Then select Save and run. The benchmark test execution starts immediately. The test result appears in the <testname> - Results page.

    To cancel a test while it is running if the results are not useful or to avoid unnecessary token usage, select Cancel test. Once canceled, the test cannot be resumed; you must start a new run if needed. You can also cancel the test using the Cancel button available on the Test card from the Tests page.

    Schedule for later Select this option to schedule the test to run at a specific time. Select the date and time, then Save. The test starts automatically at the scheduled time. The test card appears with a 'Scheduled' status, along with its planned execution date and time on the Tests page.

    Test execution may take several minutes, depending on the number of prompts.

  4. View benchmark test results.