Head to head
Kev 4B scores 76.8 and d1 82.7 on the System One Index. Kev 4B is more accurate on 2 of 16 text datasets, d1 on 14. Same cases, same request, same scorer.
| Kev 4B | d1 | |
|---|---|---|
| System One Index | 76.8 | 82.7 |
| English | 78.3 | 81.3 |
| Arabic | 74.8 | 83.9 |
| Vision Index | – | – |
| Calibration error | 5.1% | 2.9% |
| Median latency | 552 ms | 188 ms |
| Cost per 1,000 calls | $0.010 | $0.015 |
| Kev 4B | d1 | |
|---|---|---|
| AG News English | 90.5 | 90.0 |
| SST-2 English | 92.5 | 95.5 |
| SST-5 English | 52.0 | 56.5 |
| DAIR Emotion English | 54.5 | 63.5 |
| Banking77 English | 88.0 | 89.0 |
| BoolQ English | 91.5 | 90.0 |
| XNLI English English | 90.7 | 91.3 |
| Typed decisions English | 66.3 | 74.5 |
| XNLI French French | 79.3 | 84.7 |
| XNLI Arabic Arabic | 72.7 | 80.0 |
| MASSIVE scenario Arabic | 57.0 | 73.0 |
| MASSIVE intent Arabic | 76.5 | 90.0 |
| OSACT4 offensive Arabic | 88.5 | 95.5 |
| Tweet sentiment Arabic | 63.0 | 67.5 |
| MIRACL rerank Arabic | 79.5 | 90.5 |
| MIRACL pairwise Arabic | 86.5 | 91.0 |
Model pages: Kev 4B · d1. Latency for API models is the median per call through OpenRouter, timed from one machine.