Liquid AI · Open weights, local
d1-omni-600M by Liquid AI scores 65.2 on the System One Index (19th of 24). Calibration error 9.8%.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 87.5 | 17th of 24 | 92.5 | 6.5% |
| SST-2 English | 86.5 | 20th of 24 | 97.0 | 5.7% |
| SST-5 English | 47.0 | 17th of 24 | 62.5 | 11.1% |
| DAIR Emotion English | 85.0 | 2nd of 24 | 91.5 | 11.6% |
| Banking77 English | 84.5 | 13th of 24 | 98.5 | 4.0% |
| BoolQ English | 74.0 | 22nd of 24 | 92.5 | 13.1% |
| XNLI English English | 82.0 | 18th of 24 | 93.3 | 11.6% |
| Typed decisions English | 57.5 | 18th of 24 | 79.0 | 5.7% |
| XNLI French French | 65.3 | 21st of 24 | 90.7 | 15.8% |
| XNLI Arabic Arabic | 52.7 | 24th of 26 | 89.3 | 13.9% |
| MASSIVE scenario Arabic | 37.5 | 26th of 26 | 89.5 | 29.1% |
| MASSIVE intent Arabic | 52.0 | 23rd of 26 | 91.5 | 25.1% |
| OSACT4 offensive Arabic | 77.5 | 24th of 26 | 95.5 | 13.2% |
| Tweet sentiment Arabic | 32.5 | 26th of 26 | 80.0 | 44.1% |
| MIRACL rerank Arabic | 55.0 | 22nd of 26 | 90.5 | 10.7% |
| MIRACL pairwise Arabic | 66.5 | 25th of 26 | 91.0 | 7.1% |
LiquidAI/d1-omni-600MOpen-weight 600M version of d1, text only (no mmproj). LFM Open License.
Ranks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.