Head to head
Drex v2.0 scores 78.6 and Jev 1.13 79.0 on the System One Index. Drex v2.0 is more accurate on 8 of 16 text datasets, Jev 1.13 on 8. Same cases, same request, same scorer.
| Drex v2.0 | Jev 1.13 | |
|---|---|---|
| System One Index | 78.6 | 79.0 |
| English | 77.8 | 78.5 |
| Arabic | 79.6 | 79.9 |
| Vision Index | – | 83.9 |
| Calibration error | 1.5% | 6.1% |
| Median latency | 294 ms | 150 ms |
| Cost per 1,000 calls | $0.015 | $0.028 |
| Drex v2.0 | Jev 1.13 | |
|---|---|---|
| AG News English | 88.0 | 88.5 |
| SST-2 English | 94.0 | 94.5 |
| SST-5 English | 56.0 | 62.0 |
| DAIR Emotion English | 55.5 | 56.0 |
| Banking77 English | 80.5 | 79.0 |
| BoolQ English | 92.5 | 89.5 |
| XNLI English English | 87.3 | 86.0 |
| Typed decisions English | 68.7 | 72.7 |
| XNLI French French | 77.3 | 76.7 |
| XNLI Arabic Arabic | 71.3 | 69.3 |
| MASSIVE scenario Arabic | 63.5 | 63.0 |
| MASSIVE intent Arabic | 82.5 | 84.0 |
| OSACT4 offensive Arabic | 94.5 | 90.0 |
| Tweet sentiment Arabic | 69.5 | 76.0 |
| MIRACL rerank Arabic | 87.0 | 86.5 |
| MIRACL pairwise Arabic | 89.0 | 90.5 |
| Drex v2.0 | Jev 1.13 | |
|---|---|---|
| SST-2, as an image | – | 94.5 |
| Arabic sentiment, as an image | – | 59.0 |
| Oxford-IIIT Pet | – | 83.5 |
| VQAv2 yes/no | – | 88.0 |
| CountBench | – | 94.5 |
Model pages: Drex v2.0 · Jev 1.13. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.