Head to head
d1 scores 82.7 and Drex v2.0 78.6 on the System One Index. d1 is more accurate on 14 of 16 text datasets, Drex v2.0 on 2. Same cases, same request, same scorer.
| d1 | Drex v2.0 | |
|---|---|---|
| System One Index | 82.7 | 78.6 |
| English | 81.3 | 77.8 |
| Arabic | 83.9 | 79.6 |
| Vision Index | – | – |
| Calibration error | 2.9% | 1.5% |
| Median latency | 188 ms | 294 ms |
| Cost per 1,000 calls | $0.015 | $0.015 |
| d1 | Drex v2.0 | |
|---|---|---|
| AG News English | 90.0 | 88.0 |
| SST-2 English | 95.5 | 94.0 |
| SST-5 English | 56.5 | 56.0 |
| DAIR Emotion English | 63.5 | 55.5 |
| Banking77 English | 89.0 | 80.5 |
| BoolQ English | 90.0 | 92.5 |
| XNLI English English | 91.3 | 87.3 |
| Typed decisions English | 74.5 | 68.7 |
| XNLI French French | 84.7 | 77.3 |
| XNLI Arabic Arabic | 80.0 | 71.3 |
| MASSIVE scenario Arabic | 73.0 | 63.5 |
| MASSIVE intent Arabic | 90.0 | 82.5 |
| OSACT4 offensive Arabic | 95.5 | 94.5 |
| Tweet sentiment Arabic | 67.5 | 69.5 |
| MIRACL rerank Arabic | 90.5 | 87.0 |
| MIRACL pairwise Arabic | 91.0 | 89.0 |
Model pages: d1 · Drex v2.0. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.