Head to head
d1-3B scores 78.4 and Drex v2.0 78.6 on the System One Index. d1-3B is more accurate on 8 of 16 text datasets, Drex v2.0 on 6, and 2 are tied. Same cases, same request, same scorer.
| d1-3B | Drex v2.0 | |
|---|---|---|
| System One Index | 78.4 | 78.6 |
| English | 78.4 | 77.8 |
| Arabic | 78.3 | 79.6 |
| Vision Index | – | – |
| Calibration error | 2.4% | 1.5% |
| Median latency | – | 294 ms |
| Cost per 1,000 calls | – | $0.015 |
| d1-3B | Drex v2.0 | |
|---|---|---|
| AG News English | 88.5 | 88.0 |
| SST-2 English | 95.5 | 94.0 |
| SST-5 English | 56.0 | 56.0 |
| DAIR Emotion English | 63.5 | 55.5 |
| Banking77 English | 88.0 | 80.5 |
| BoolQ English | 85.0 | 92.5 |
| XNLI English English | 86.0 | 87.3 |
| Typed decisions English | 64.8 | 68.7 |
| XNLI French French | 78.7 | 77.3 |
| XNLI Arabic Arabic | 73.3 | 71.3 |
| MASSIVE scenario Arabic | 71.5 | 63.5 |
| MASSIVE intent Arabic | 83.5 | 82.5 |
| OSACT4 offensive Arabic | 84.5 | 94.5 |
| Tweet sentiment Arabic | 63.5 | 69.5 |
| MIRACL rerank Arabic | 82.5 | 87.0 |
| MIRACL pairwise Arabic | 89.0 | 89.0 |
Model pages: d1-3B · Drex v2.0. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.