Jared Palmer · Open weights, local
Kev 0.8B by Jared Palmer scores 63.6 on the System One Index (20th of 24). Calibration error 4.3%.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 91.0 | 5th of 24 | 92.5 | 4.0% |
| SST-2 English | 90.0 | 17th of 24 | 97.0 | 4.6% |
| SST-5 English | 47.0 | 17th of 24 | 62.5 | 16.2% |
| DAIR Emotion English | 56.0 | 17th of 24 | 91.5 | 8.6% |
| Banking77 English | 80.5 | 15th of 24 | 98.5 | 11.4% |
| BoolQ English | 76.0 | 21st of 24 | 92.5 | 12.0% |
| XNLI English English | 81.3 | 19th of 24 | 93.3 | 3.8% |
| Typed decisions English | 46.3 | 22nd of 24 | 79.0 | 4.6% |
| XNLI French French | 66.0 | 19th of 24 | 90.7 | 12.7% |
| XNLI Arabic Arabic | 58.7 | 20th of 26 | 89.3 | 18.6% |
| MASSIVE scenario Arabic | 38.5 | 25th of 26 | 89.5 | 12.8% |
| MASSIVE intent Arabic | 51.5 | 25th of 26 | 91.5 | 8.4% |
| OSACT4 offensive Arabic | 49.0 | 26th of 26 | 95.5 | 15.1% |
| Tweet sentiment Arabic | 57.0 | 19th of 26 | 80.0 | 10.8% |
| MIRACL rerank Arabic | 62.5 | 21st of 26 | 90.5 | 24.8% |
| MIRACL pairwise Arabic | 67.0 | 24th of 26 | 91.0 | 4.4% |
jaredpalmer/kev-0.8bEnglish-only, like Kev 4B. Apache 2.0.
Ranks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.