~/system-one-bench / compare/patrick-18-dexy-decision-9b-vs-vllm-semantic-router-d3-nano
Head to head
Dexy-Decision-9B scores 84.7 and d3 nano 78.0 on the System One Index. Dexy-Decision-9B is more accurate on 13 of 16 text datasets, d3 nano on 2, and 1 are tied. Same cases, same request, same scorer.
| Dexy-Decision-9B | d3 nano | |
|---|---|---|
| System One Index | 84.7 | 78.0 |
| English | 82.3 | 77.9 |
| Arabic | 86.8 | 76.9 |
| Vision Index | – | – |
| Calibration error | 3.5% | 7.0% |
| Median latency | – | – |
| Cost per 1,000 calls | – | – |
| Dexy-Decision-9B | d3 nano | |
|---|---|---|
| AG News English | 91.5 | 88.0 |
| SST-2 English | 95.5 | 89.5 |
| SST-5 English | 60.0 | 55.5 |
| DAIR Emotion English | 54.0 | 59.5 |
| Banking77 English | 93.0 | 93.0 |
| BoolQ English | 89.5 | 88.0 |
| XNLI English English | 94.7 | 87.3 |
| Typed decisions English | 80.0 | 62.6 |
| XNLI French French | 89.3 | 86.0 |
| XNLI Arabic Arabic | 89.3 | 82.0 |
| MASSIVE scenario Arabic | 89.5 | 61.5 |
| MASSIVE intent Arabic | 91.5 | 81.0 |
| OSACT4 offensive Arabic | 91.5 | 79.0 |
| Tweet sentiment Arabic | 67.0 | 71.0 |
| MIRACL rerank Arabic | 88.0 | 77.5 |
| MIRACL pairwise Arabic | 91.0 | 86.5 |
Model pages: Dexy-Decision-9B · d3 nano. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.