Skip to main content
Once your evaluation job has finished, go to Evaluations and click on your evaluation to view the results. Each evaluator’s score indicates how well the model performed against the defined criteria. Clicking on Explore Results gives access to individual sample-level results.

​
RESULTS FOR TOOL-USING CONVERSATIONS

When the evaluation attached tools or scored traces that contain tool calls, also check the following:
  • Tools section: Lists the tools that were available to the job that generated the scored responses, with the tool implementations that served each one. A run with inference lists the tools it offered the model. A run that scored existing conversations lists the tools recorded with the dataset version. In the API, this list is the evaluation’s tools field.
  • Sample table: The Turns column shows the number of turns in each conversation, and the Used tools column shows the tools the conversation called and how many times.
  • Failure modes: Failure mode analysis reads the failed conversations, including their tool calls. See Diagnosing failure modes.

​
RESULTS FOR RUBRIC EVALUATORS

A rubric evaluator reports both an overall score and a score for each criterion:
  • Overall score: Combined the way the rubric’s aggregation says, such as a weighted mean of four criteria. The unweighted mean appears next to it, so you can see how much the weights change the result.
  • Criteria: Expand the rubric’s row in the evaluators table to see each criterion’s weight, the labels the judge returned, and its average score.
  • Samples: Each sample shows the label, score, weight, and rationale for every criterion.
A criterion’s average counts only the rows where it received a scored label. Rows labeled with an unscored value, such as not_applicable, do not lower it.

​
INTERPRETING RESULTS

You can use your evaluation results to:
  • Compare baseline models to fine-tuned models
  • Identify regressions or improvements due after model changes
  • Decide whether to retrain, adjust data, or refine evaluators
Evaluation results become significantly more valuable when paired with failure mode analysis and data synthesis. By identifying where and why a model underperforms, you can systematically generate targeted data to address those weaknesses, creating a tight feedback loop between evaluation, diagnosis, and improvement. This integrated approach enables more efficient iteration and drives measurable gains in model performance.