vLLM Semantic Router · Open weights, local
d3 edge by vLLM Semantic Router scores 65.9 on the System One Index (18th of 24). Calibration error 18.7%.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 80.0 | 22nd of 24 | 92.5 | 9.9% |
| SST-2 English | 80.5 | 22nd of 24 | 97.0 | 8.5% |
| SST-5 English | 39.5 | 21st of 24 | 62.5 | 12.4% |
| DAIR Emotion English | 57.0 | 12th of 24 | 91.5 | 20.8% |
| Banking77 English | 90.0 | 6th of 24 | 98.5 | 1.8% |
| BoolQ English | 79.5 | 20th of 24 | 92.5 | 5.8% |
| XNLI English English | 84.0 | 16th of 24 | 93.3 | 3.6% |
| Typed decisions English | 42.3 | 23rd of 24 | 79.0 | 34.4% |
| XNLI French French | 66.0 | 19th of 24 | 90.7 | 8.4% |
| XNLI Arabic Arabic | 59.3 | 19th of 26 | 89.3 | 12.2% |
| MASSIVE scenario Arabic | 41.0 | 23rd of 26 | 89.5 | 21.1% |
| MASSIVE intent Arabic | 55.0 | 21st of 26 | 91.5 | 10.6% |
| OSACT4 offensive Arabic | 72.5 | 25th of 26 | 95.5 | 2.9% |
| Tweet sentiment Arabic | 61.5 | 16th of 26 | 80.0 | 4.3% |
| MIRACL rerank Arabic | 70.0 | 20th of 26 | 90.5 | 5.6% |
| MIRACL pairwise Arabic | 76.0 | 21st of 26 | 91.0 | 5.0% |
vllm-sr/d3-edgeDecision 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.