Head to head
Drex v1.5 scores 77.6 and d3 edge 65.9 on the System One Index. Drex v1.5 is more accurate on 14 of 16 text datasets, d3 edge on 2. Same cases, same request, same scorer.
| Drex v1.5 | d3 edge | |
|---|---|---|
| System One Index | 77.6 | 65.9 |
| English | 77.6 | 69.1 |
| Arabic | 77.0 | 62.2 |
| Vision Index | – | – |
| Calibration error | 10.8% | 18.7% |
| Median latency | 198 ms | – |
| Cost per 1,000 calls | $0.010 | – |
| Drex v1.5 | d3 edge | |
|---|---|---|
| AG News English | 90.0 | 80.0 |
| SST-2 English | 94.5 | 80.5 |
| SST-5 English | 53.5 | 39.5 |
| DAIR Emotion English | 53.5 | 57.0 |
| Banking77 English | 87.0 | 90.0 |
| BoolQ English | 89.0 | 79.5 |
| XNLI English English | 88.7 | 84.0 |
| Typed decisions English | 64.8 | 42.3 |
| XNLI French French | 82.0 | 66.0 |
| XNLI Arabic Arabic | 74.0 | 59.3 |
| MASSIVE scenario Arabic | 58.5 | 41.0 |
| MASSIVE intent Arabic | 77.5 | 55.0 |
| OSACT4 offensive Arabic | 90.5 | 72.5 |
| Tweet sentiment Arabic | 63.5 | 61.5 |
| MIRACL rerank Arabic | 89.0 | 70.0 |
| MIRACL pairwise Arabic | 86.0 | 76.0 |
Model pages: Drex v1.5 · d3 edge. Latency for API models is the median per call through OpenRouter, timed from one machine.