Nace AI · OpenRouter API
Drex v1.5 by Nace AI scores 77.6 on the System One Index (12th of 24). Calibration error 10.8%, median latency 198 ms.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 90.0 | 8th of 24 | 92.5 | 7.0% |
| SST-2 English | 94.5 | 10th of 24 | 97.0 | 2.5% |
| SST-5 English | 53.5 | 10th of 24 | 62.5 | 10.3% |
| DAIR Emotion English | 53.5 | 23rd of 24 | 91.5 | 24.2% |
| Banking77 English | 87.0 | 10th of 24 | 98.5 | 6.7% |
| BoolQ English | 89.0 | 10th of 24 | 92.5 | 6.6% |
| XNLI English English | 88.7 | 7th of 24 | 93.3 | 3.1% |
| Typed decisions English | 64.8 | 15th of 24 | 79.0 | 13.6% |
| XNLI French French | 82.0 | 7th of 24 | 90.7 | 10.1% |
| XNLI Arabic Arabic | 74.0 | 8th of 26 | 89.3 | 14.0% |
| MASSIVE scenario Arabic | 58.5 | 16th of 26 | 89.5 | 25.0% |
| MASSIVE intent Arabic | 77.5 | 13th of 26 | 91.5 | 11.5% |
| OSACT4 offensive Arabic | 90.5 | 5th of 26 | 95.5 | 3.3% |
| Tweet sentiment Arabic | 63.5 | 12th of 26 | 80.0 | 17.8% |
| MIRACL rerank Arabic | 89.0 | 3rd of 26 | 90.5 | 5.3% |
| MIRACL pairwise Arabic | 86.0 | 16th of 26 | 91.0 | 9.1% |
nace-ai/drex-v1.5Ranks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.