Head to head
Jev 1.13 scores 79.0 and Jev 1.14 79.3 on the System One Index. Jev 1.13 is more accurate on 4 of 16 text datasets, Jev 1.14 on 9, and 3 are tied. Same cases, same request, same scorer.
| Jev 1.13 | Jev 1.14 | |
|---|---|---|
| System One Index | 79.0 | 79.3 |
| English | 78.5 | 78.8 |
| Arabic | 79.9 | 80.3 |
| Vision Index | 83.9 | 85.1 |
| Calibration error | 6.1% | 6.0% |
| Median latency | 150 ms | 136 ms |
| Cost per 1,000 calls | $0.028 | $0.028 |
| Jev 1.13 | Jev 1.14 | |
|---|---|---|
| AG News English | 88.5 | 89.0 |
| SST-2 English | 94.5 | 95.0 |
| SST-5 English | 62.0 | 60.0 |
| DAIR Emotion English | 56.0 | 57.5 |
| Banking77 English | 79.0 | 78.5 |
| BoolQ English | 89.5 | 90.0 |
| XNLI English English | 86.0 | 86.0 |
| Typed decisions English | 72.7 | 74.2 |
| XNLI French French | 76.7 | 76.7 |
| XNLI Arabic Arabic | 69.3 | 71.3 |
| MASSIVE scenario Arabic | 63.0 | 65.0 |
| MASSIVE intent Arabic | 84.0 | 85.0 |
| OSACT4 offensive Arabic | 90.0 | 87.0 |
| Tweet sentiment Arabic | 76.0 | 76.0 |
| MIRACL rerank Arabic | 86.5 | 87.5 |
| MIRACL pairwise Arabic | 90.5 | 90.0 |
| Jev 1.13 | Jev 1.14 | |
|---|---|---|
| SST-2, as an image | 94.5 | 93.5 |
| Arabic sentiment, as an image | 59.0 | 63.5 |
| Oxford-IIIT Pet | 83.5 | 88.5 |
| VQAv2 yes/no | 88.0 | 86.5 |
| CountBench | 94.5 | 93.5 |
Model pages: Jev 1.13 ยท Jev 1.14. Latency for API models is the median per call through OpenRouter, timed from one machine.