View benchmark test results

Test execution time depends on the number of prompts included in the benchmark. After execution starts, the <test name> - Results page displays the following information.

Procedure
  1. Execute a benchmark test by running immediately or scheduling later.
  2. Once the benchmark test is executed based on the selected option, system does either of the following:
    • For immediate test execution, the <testname> - Results page displays the following.

      Result

      Description

      Test status

      Displays the current execution state as Running, Completed, Suspended, Scheduled, or Cancelled.

      Summary metrics

      Displays high-level benchmark statistics, that include:

      • Average score (0-5 scale)
      • Total prompts executed.
      • Average response time per prompt.
      • Success rate.

      Result filters

      Use the filter tabs to quickly review benchmark results by score or response time.

      The comparison results are displayed in a tabular format for easier analysis and comparison.

      All Displays all prompt results included in the benchmark run.
      Score 5 Displays only perfect responses with a score of 5.
      Score 4 Displays high-quality responses with a score of 4.
      Score <3 Displays low-performing responses with a score equal to or below 3.
      Slow (>3s) Displays responses that took longer than 3 seconds to generate.
      Result 1 higher Displays if result 1 is better than result 2.
      Result 2 higher Displays if result 2 is better than result 1.

      Detailed prompt evaluations

      Displays prompt-level benchmark details, that include:

      • Expected answer

      • Actual generated response

      • Reference match validation

        Reference matching verifies whether the sources used in the generated response match the expected references defined for the benchmark test.

      • Set as reference answer option. When a generated response is considered more accurate, complete, or appropriate than the current expected result, the selected response becomes the new reference answer used for future evaluations, and this is recorded in the audit log.

      AI evaluation

      Displays AI-generated evaluation comments and scoring for benchmark results.

    • For test execution scheduled at a later time, a test card appears on the Tests page with the status as Scheduled.

  3. To exit the results page, select Close in the top-right corner.

    The Tests page appears.

  4. Each benchmark test appears as a test card containing execution details.

    When a benchmark test is running, the status is displayed as Running. Each test card includes the following information:

    Status Description
    Provider Generative AI provider used for the test, such as AI Tungsten.
    Prompts Number of prompts included in the benchmark.
    Test status Current test status with timestamp.
    Progress Number of prompts executed so far (running tests only) .
    Duration Total execution time for completed tests.
    Performance rating Overall benchmark score on a 0-5 scale.
    Latest results Opens the most recent benchmark results.
    View runs Displays execution history for a benchmark test. It helps review previous runs, compare performance across executions, delete test runs, and access detailed results for individual test runs. The list is ordered chronologically, with the most recent run displayed at the top.
    Compare Compare two benchmark test runs.
    Cancel Stops a currently running test.
    Re-run To rerun a suspended test.
    View error Lists the error details for the suspended test.

    Re-run and View error options are available only for a suspended test.