Head to head
Kev 4B scores 76.8 and Tev1 4B Experimental 71.5 on the System One Index. Kev 4B is more accurate on 7 of 16 text datasets, Tev1 4B Experimental on 9. Same cases, same request, same scorer.
| Kev 4B | Tev1 4B Experimental | |
|---|---|---|
| System One Index | 76.8 | 71.5 |
| English | 78.3 | 66.4 |
| Arabic | 74.8 | 75.8 |
| Vision Index | – | – |
| Calibration error | 5.1% | 5.3% |
| Median latency | 552 ms | 232 ms |
| Cost per 1,000 calls | $0.010 | $0.019 |
| Kev 4B | Tev1 4B Experimental | |
|---|---|---|
| AG News English | 90.5 | 91.5 |
| SST-2 English | 92.5 | 94.0 |
| SST-5 English | 52.0 | 43.0 |
| DAIR Emotion English | 54.5 | 56.5 |
| Banking77 English | 88.0 | 1.0 |
| BoolQ English | 91.5 | 91.0 |
| XNLI English English | 90.7 | 87.3 |
| Typed decisions English | 66.3 | 67.0 |
| XNLI French French | 79.3 | 82.7 |
| XNLI Arabic Arabic | 72.7 | 76.0 |
| MASSIVE scenario Arabic | 57.0 | 56.5 |
| MASSIVE intent Arabic | 76.5 | 70.0 |
| OSACT4 offensive Arabic | 88.5 | 84.5 |
| Tweet sentiment Arabic | 63.0 | 70.5 |
| MIRACL rerank Arabic | 79.5 | 85.5 |
| MIRACL pairwise Arabic | 86.5 | 87.5 |
Model pages: Kev 4B · Tev1 4B Experimental. Latency for API models is the median per call through OpenRouter, timed from one machine.