vLLM Semantic Router · Open weights, local
d3 nano by vLLM Semantic Router scores 78.0 on the System One Index (11th of 24). Calibration error 7.0%.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 88.0 | 16th of 24 | 92.5 | 10.1% |
| SST-2 English | 89.5 | 18th of 24 | 97.0 | 3.1% |
| SST-5 English | 55.5 | 9th of 24 | 62.5 | 6.2% |
| DAIR Emotion English | 59.5 | 8th of 24 | 91.5 | 16.4% |
| Banking77 English | 93.0 | 4th of 24 | 98.5 | 5.0% |
| BoolQ English | 88.0 | 13th of 24 | 92.5 | 4.5% |
| XNLI English English | 87.3 | 9th of 24 | 93.3 | 5.4% |
| Typed decisions English | 62.6 | 16th of 24 | 79.0 | 11.5% |
| XNLI French French | 86.0 | 2nd of 24 | 90.7 | 4.8% |
| XNLI Arabic Arabic | 82.0 | 2nd of 26 | 89.3 | 6.9% |
| MASSIVE scenario Arabic | 61.5 | 14th of 26 | 89.5 | 16.2% |
| MASSIVE intent Arabic | 81.0 | 12th of 26 | 91.5 | 8.0% |
| OSACT4 offensive Arabic | 79.0 | 23rd of 26 | 95.5 | 3.9% |
| Tweet sentiment Arabic | 71.0 | 6th of 26 | 80.0 | 6.1% |
| MIRACL rerank Arabic | 77.5 | 17th of 26 | 90.5 | 8.4% |
| MIRACL pairwise Arabic | 86.5 | 12th of 26 | 91.0 | 4.4% |
vllm-sr/d3-nanoDecision 3.0, text only (no mmproj). 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.