Head to head
Naluri-279M scores 50.4 and Naluri-308M 61.0 on the System One Index. Naluri-279M is more accurate on 4 of 16 text datasets, Naluri-308M on 12. Same cases, same request, same scorer.
| Naluri-279M | Naluri-308M | |
|---|---|---|
| System One Index | 50.4 | 61.0 |
| English | 52.6 | 59.3 |
| Arabic | 50.7 | 61.9 |
| Vision Index | – | – |
| Calibration error | 8.3% | 5.7% |
| Median latency | – | – |
| Cost per 1,000 calls | – | – |
| Naluri-279M | Naluri-308M | |
|---|---|---|
| AG News English | 77.5 | 79.5 |
| SST-2 English | 77.0 | 68.5 |
| SST-5 English | 34.0 | 28.0 |
| DAIR Emotion English | 68.5 | 84.0 |
| Banking77 English | 31.0 | 27.0 |
| BoolQ English | 54.5 | 56.5 |
| XNLI English English | 30.7 | 76.0 |
| Typed decisions English | 47.3 | 54.7 |
| XNLI French French | 31.3 | 68.0 |
| XNLI Arabic Arabic | 30.7 | 56.7 |
| MASSIVE scenario Arabic | 58.0 | 63.5 |
| MASSIVE intent Arabic | 52.0 | 65.5 |
| OSACT4 offensive Arabic | 82.5 | 79.5 |
| Tweet sentiment Arabic | 53.0 | 58.0 |
| MIRACL rerank Arabic | 28.0 | 34.0 |
| MIRACL pairwise Arabic | 50.5 | 76.0 |
Model pages: Naluri-279M · Naluri-308M. Latency for API models is the median per call through OpenRouter, timed from one machine.