Nace AI · OpenRouter API
Drex v2.0 by Nace AI scores 78.6 on the System One Index (14th of 30). Calibration error 1.5%, median latency 294 ms.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 88.0 | 19th of 30 | 92.5 | 4.3% |
| SST-2 English | 94.0 | 16th of 30 | 97.0 | 11.7% |
| SST-5 English | 56.0 | 11th of 30 | 62.5 | 8.7% |
| DAIR Emotion English | 55.5 | 25th of 30 | 91.5 | 15.2% |
| Banking77 English | 80.5 | 19th of 30 | 98.5 | 3.2% |
| BoolQ English | 92.5 | 2nd of 30 | 93.0 | 3.3% |
| XNLI English English | 87.3 | 12th of 30 | 95.3 | 6.3% |
| Typed decisions English | 68.7 | 15th of 30 | 79.0 | 4.5% |
| XNLI French French | 77.3 | 17th of 30 | 91.3 | 7.1% |
| XNLI Arabic Arabic | 71.3 | 17th of 32 | 90.0 | 6.2% |
| MASSIVE scenario Arabic | 63.5 | 16th of 32 | 89.5 | 9.7% |
| MASSIVE intent Arabic | 82.5 | 16th of 32 | 91.5 | 8.1% |
| OSACT4 offensive Arabic | 94.5 | 4th of 32 | 97.0 | 4.7% |
| Tweet sentiment Arabic | 69.5 | 11th of 32 | 80.0 | 5.7% |
| MIRACL rerank Arabic | 87.0 | 12th of 32 | 90.5 | 4.5% |
| MIRACL pairwise Arabic | 89.0 | 12th of 32 | 91.5 | 2.9% |
nace-ai/drex-v2.035B mixture-of-experts, 3B active. Text only.
Ranks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.