Head to head
Naluri-308M scores 61.0 and Drex v1.5 77.6 on the System One Index. Naluri-308M is more accurate on 2 of 16 text datasets, Drex v1.5 on 14. Same cases, same request, same scorer.
| Naluri-308M | Drex v1.5 | |
|---|---|---|
| System One Index | 61.0 | 77.6 |
| English | 59.3 | 77.6 |
| Arabic | 61.9 | 77.0 |
| Vision Index | – | – |
| Calibration error | 5.7% | 10.8% |
| Median latency | – | 198 ms |
| Cost per 1,000 calls | – | $0.010 |
| Naluri-308M | Drex v1.5 | |
|---|---|---|
| AG News English | 79.5 | 90.0 |
| SST-2 English | 68.5 | 94.5 |
| SST-5 English | 28.0 | 53.5 |
| DAIR Emotion English | 84.0 | 53.5 |
| Banking77 English | 27.0 | 87.0 |
| BoolQ English | 56.5 | 89.0 |
| XNLI English English | 76.0 | 88.7 |
| Typed decisions English | 54.7 | 64.8 |
| XNLI French French | 68.0 | 82.0 |
| XNLI Arabic Arabic | 56.7 | 74.0 |
| MASSIVE scenario Arabic | 63.5 | 58.5 |
| MASSIVE intent Arabic | 65.5 | 77.5 |
| OSACT4 offensive Arabic | 79.5 | 90.5 |
| Tweet sentiment Arabic | 58.0 | 63.5 |
| MIRACL rerank Arabic | 34.0 | 89.0 |
| MIRACL pairwise Arabic | 76.0 | 86.0 |
Model pages: Naluri-308M · Drex v1.5. Latency for API models is the median per call through OpenRouter, timed from one machine.