TypeSafe ยท OpenRouter API
Jev 1.14 by TypeSafe scores 79.3 on the System One Index (8th of 24) and 85.1 on the Vision Index (1st of 8). Calibration error 6.0%, median latency 136 ms.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 89.0 | 11th of 24 | 92.5 | 6.9% |
| SST-2 English | 95.0 | 8th of 24 | 97.0 | 7.7% |
| SST-5 English | 60.0 | 5th of 24 | 62.5 | 18.1% |
| DAIR Emotion English | 57.5 | 11th of 24 | 91.5 | 27.0% |
| Banking77 English | 78.5 | 17th of 24 | 98.5 | 10.4% |
| BoolQ English | 90.0 | 6th of 24 | 92.5 | 2.7% |
| XNLI English English | 86.0 | 12th of 24 | 93.3 | 7.0% |
| Typed decisions English | 74.2 | 3rd of 24 | 79.0 | 5.0% |
| XNLI French French | 76.7 | 14th of 24 | 90.7 | 10.7% |
| XNLI Arabic Arabic | 71.3 | 14th of 26 | 89.3 | 11.4% |
| MASSIVE scenario Arabic | 65.0 | 11th of 26 | 89.5 | 16.3% |
| MASSIVE intent Arabic | 85.0 | 7th of 26 | 91.5 | 2.8% |
| OSACT4 offensive Arabic | 87.0 | 13th of 26 | 95.5 | 4.8% |
| Tweet sentiment Arabic | 76.0 | 2nd of 26 | 80.0 | 9.5% |
| MIRACL rerank Arabic | 87.5 | 7th of 26 | 90.5 | 6.6% |
| MIRACL pairwise Arabic | 90.0 | 4th of 26 | 91.0 | 2.1% |
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| SST-2, as an image | 93.5 | 5th of 8 | 96.0 | 6.0% |
| Arabic sentiment, as an image | 63.5 | 2nd of 8 | 67.0 | 5.4% |
| Oxford-IIIT Pet | 88.5 | 1st of 8 | 88.5 | 5.0% |
| VQAv2 yes/no | 86.5 | 7th of 8 | 89.5 | 5.8% |
| CountBench | 93.5 | 3rd of 8 | 94.5 | 4.7% |
typesafe/jev-1.14Preview release
Ranks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.