Head to head
Kev 0.8B scores 63.6 and Drex v1.5 77.6 on the System One Index. Kev 0.8B is more accurate on 2 of 16 text datasets, Drex v1.5 on 14. Same cases, same request, same scorer.
| Kev 0.8B | Drex v1.5 | |
|---|---|---|
| System One Index | 63.6 | 77.6 |
| English | 71.0 | 77.6 |
| Arabic | 54.9 | 77.0 |
| Vision Index | – | – |
| Calibration error | 4.3% | 10.8% |
| Median latency | – | 198 ms |
| Cost per 1,000 calls | – | $0.010 |
| Kev 0.8B | Drex v1.5 | |
|---|---|---|
| AG News English | 91.0 | 90.0 |
| SST-2 English | 90.0 | 94.5 |
| SST-5 English | 47.0 | 53.5 |
| DAIR Emotion English | 56.0 | 53.5 |
| Banking77 English | 80.5 | 87.0 |
| BoolQ English | 76.0 | 89.0 |
| XNLI English English | 81.3 | 88.7 |
| Typed decisions English | 46.3 | 64.8 |
| XNLI French French | 66.0 | 82.0 |
| XNLI Arabic Arabic | 58.7 | 74.0 |
| MASSIVE scenario Arabic | 38.5 | 58.5 |
| MASSIVE intent Arabic | 51.5 | 77.5 |
| OSACT4 offensive Arabic | 49.0 | 90.5 |
| Tweet sentiment Arabic | 57.0 | 63.5 |
| MIRACL rerank Arabic | 62.5 | 89.0 |
| MIRACL pairwise Arabic | 67.0 | 86.0 |
Model pages: Kev 0.8B · Drex v1.5. Latency for API models is the median per call through OpenRouter, timed from one machine.