Head to head
Drex v2.0 scores 78.6 and Tev1 4B Experimental 71.5 on the System One Index. Drex v2.0 is more accurate on 9 of 16 text datasets, Tev1 4B Experimental on 5, and 2 are tied. Same cases, same request, same scorer.
| Drex v2.0 | Tev1 4B Experimental | |
|---|---|---|
| System One Index | 78.6 | 71.5 |
| English | 77.8 | 66.4 |
| Arabic | 79.6 | 75.8 |
| Vision Index | – | – |
| Calibration error | 1.5% | 5.3% |
| Median latency | 294 ms | 232 ms |
| Cost per 1,000 calls | $0.015 | $0.019 |
| Drex v2.0 | Tev1 4B Experimental | |
|---|---|---|
| AG News English | 88.0 | 91.5 |
| SST-2 English | 94.0 | 94.0 |
| SST-5 English | 56.0 | 43.0 |
| DAIR Emotion English | 55.5 | 56.5 |
| Banking77 English | 80.5 | 1.0 |
| BoolQ English | 92.5 | 91.0 |
| XNLI English English | 87.3 | 87.3 |
| Typed decisions English | 68.7 | 67.0 |
| XNLI French French | 77.3 | 82.7 |
| XNLI Arabic Arabic | 71.3 | 76.0 |
| MASSIVE scenario Arabic | 63.5 | 56.5 |
| MASSIVE intent Arabic | 82.5 | 70.0 |
| OSACT4 offensive Arabic | 94.5 | 84.5 |
| Tweet sentiment Arabic | 69.5 | 70.5 |
| MIRACL rerank Arabic | 87.0 | 85.5 |
| MIRACL pairwise Arabic | 89.0 | 87.5 |
Model pages: Drex v2.0 · Tev1 4B Experimental. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.