Head to head
Tev1 4B Experimental scores 71.5 and d3 nano 78.0 on the System One Index. Tev1 4B Experimental is more accurate on 7 of 16 text datasets, d3 nano on 8, and 1 are tied. Same cases, same request, same scorer.
| Tev1 4B Experimental | d3 nano | |
|---|---|---|
| System One Index | 71.5 | 78.0 |
| English | 66.4 | 77.9 |
| Arabic | 75.8 | 76.9 |
| Vision Index | – | – |
| Calibration error | 5.3% | 7.0% |
| Median latency | 232 ms | – |
| Cost per 1,000 calls | $0.019 | – |
| Tev1 4B Experimental | d3 nano | |
|---|---|---|
| AG News English | 91.5 | 88.0 |
| SST-2 English | 94.0 | 89.5 |
| SST-5 English | 43.0 | 55.5 |
| DAIR Emotion English | 56.5 | 59.5 |
| Banking77 English | 1.0 | 93.0 |
| BoolQ English | 91.0 | 88.0 |
| XNLI English English | 87.3 | 87.3 |
| Typed decisions English | 67.0 | 62.6 |
| XNLI French French | 82.7 | 86.0 |
| XNLI Arabic Arabic | 76.0 | 82.0 |
| MASSIVE scenario Arabic | 56.5 | 61.5 |
| MASSIVE intent Arabic | 70.0 | 81.0 |
| OSACT4 offensive Arabic | 84.5 | 79.0 |
| Tweet sentiment Arabic | 70.5 | 71.0 |
| MIRACL rerank Arabic | 85.5 | 77.5 |
| MIRACL pairwise Arabic | 87.5 | 86.5 |
Model pages: Tev1 4B Experimental · d3 nano. Latency for API models is the median per call through OpenRouter, timed from one machine.