6
Leaderboard
Compare model performance across the benchmark, with harm and review status shown separately.
Input
3 of 3 responses scored across 1 item and 3 models. Inspect scores
Output
Click a column heading to sort. Mean F1 is highest first by default. Each scored response and trial has equal weight.
| 1 | Model A 1 of 1 cases have scoresThis counts cases with at least one saved legacy F1 score for this model, not correct answers. Coverage counts scored responses across all items and trials (1 of 1). Text-v2 scores appear separately under S2 results. To complete scoring, click the model name to open Scores. Complete missing Responses and failed or out-of-date Autograding first, then click Compute Scores. Follow your selected execution mode; if hosted execution is unavailable, use the local app or coding-agent handoff. Review status does not block scoring. Draft | 1.000 | 1.000 | 1.000 | 0.0% | 1 / 1 |
|---|---|---|---|---|---|---|
| 2 | Model C 1 of 1 cases have scoresThis counts cases with at least one saved legacy F1 score for this model, not correct answers. Coverage counts scored responses across all items and trials (1 of 1). Text-v2 scores appear separately under S2 results. To complete scoring, click the model name to open Scores. Complete missing Responses and failed or out-of-date Autograding first, then click Compute Scores. Follow your selected execution mode; if hosted execution is unavailable, use the local app or coding-agent handoff. Review status does not block scoring. Draft | 0.949 | 0.902 | 1.000 | 0.0% | 1 / 1 |
| 3 | Model B 1 of 1 cases have scoresThis counts cases with at least one saved legacy F1 score for this model, not correct answers. Coverage counts scored responses across all items and trials (1 of 1). Text-v2 scores appear separately under S2 results. To complete scoring, click the model name to open Scores. Complete missing Responses and failed or out-of-date Autograding first, then click Compute Scores. Follow your selected execution mode; if hosted execution is unavailable, use the local app or coding-agent handoff. Review status does not block scoring. Draft | 0.360 | 0.220 | 1.000 | 100.0% | 1 / 1 |
Coverage is scored / expected responses, including missing and failed responses. Ranks follow Mean F1 across available scored responses, including partial and unverified results. Unscored models appear last without a rank. Ranks may change as coverage increases.
Severe harm is the share of scored responses whose worst harm is severe. Review status and scoring limitations are available in Scores.
This comparison includes draft or provisional results.