Head to head
Kev 4B scores 76.8 and d3 edge 65.9 on the System One Index. Kev 4B is more accurate on 14 of 16 text datasets, d3 edge on 2. Same cases, same request, same scorer.
| Kev 4B | d3 edge | |
|---|---|---|
| System One Index | 76.8 | 65.9 |
| English | 78.3 | 69.1 |
| Arabic | 74.8 | 62.2 |
| Vision Index | – | – |
| Calibration error | 5.1% | 18.7% |
| Median latency | 552 ms | – |
| Cost per 1,000 calls | $0.010 | – |
| Kev 4B | d3 edge | |
|---|---|---|
| AG News English | 90.5 | 80.0 |
| SST-2 English | 92.5 | 80.5 |
| SST-5 English | 52.0 | 39.5 |
| DAIR Emotion English | 54.5 | 57.0 |
| Banking77 English | 88.0 | 90.0 |
| BoolQ English | 91.5 | 79.5 |
| XNLI English English | 90.7 | 84.0 |
| Typed decisions English | 66.3 | 42.3 |
| XNLI French French | 79.3 | 66.0 |
| XNLI Arabic Arabic | 72.7 | 59.3 |
| MASSIVE scenario Arabic | 57.0 | 41.0 |
| MASSIVE intent Arabic | 76.5 | 55.0 |
| OSACT4 offensive Arabic | 88.5 | 72.5 |
| Tweet sentiment Arabic | 63.0 | 61.5 |
| MIRACL rerank Arabic | 79.5 | 70.0 |
| MIRACL pairwise Arabic | 86.5 | 76.0 |
Model pages: Kev 4B · d3 edge. Latency for API models is the median per call through OpenRouter, timed from one machine.