Rank Alignment Correlation on BigCodeBench Hard Set Complete
0.994Spearman's RhoConv-to-Bench
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Conv-to-BenchEvaluator=Gemini-2.5-Flash, Configuration=Instructions-Only2026.04 | 0.994 | 0.001 | 0.982 | 0.001 | |
| Arena-Hard-Auto (Code)Evaluator=Gemini-2.5-Flash2026.04 | 0.97 | 0.001 | 0.909 | 0.002 | |
| Conv-to-BenchEvaluator=GPT-5-Mini, Configuration=Instructions-Only2026.04 | 0.946 | 0.001 | 0.837 | 0.004 | |
| Arena-Hard-Auto (Code)Evaluator=GPT-5-Mini2026.04 | 0.922 | 0.001 | 0.837 | 0.004 |