Head to head
Classify-1 scores 30.6 and Tev1 4B Experimental 71.5 on the System One Index. Classify-1 is more accurate on 1 of 16 text datasets, Tev1 4B Experimental on 14, and 1 are tied. Same cases, same request, same scorer.
| Classify-1 | Tev1 4B Experimental | |
|---|---|---|
| System One Index | 30.6 | 71.5 |
| English | 30.7 | 66.4 |
| Arabic | 30.6 | 75.8 |
| Vision Index | – | – |
| Calibration error | 3.2% | 5.3% |
| Median latency | 254 ms | 232 ms |
| Cost per 1,000 calls | $0.018 | $0.019 |
| Classify-1 | Tev1 4B Experimental | |
|---|---|---|
| AG News English | 26.5 | 91.5 |
| SST-2 English | 50.0 | 94.0 |
| SST-5 English | 13.0 | 43.0 |
| DAIR Emotion English | 59.0 | 56.5 |
| Banking77 English | 1.0 | 1.0 |
| BoolQ English | 38.0 | 91.0 |
| XNLI English English | 30.7 | 87.3 |
| Typed decisions English | 27.2 | 67.0 |
| XNLI French French | 30.7 | 82.7 |
| XNLI Arabic Arabic | 30.7 | 76.0 |
| MASSIVE scenario Arabic | 3.0 | 56.5 |
| MASSIVE intent Arabic | 3.5 | 70.0 |
| OSACT4 offensive Arabic | 80.0 | 84.5 |
| Tweet sentiment Arabic | 35.0 | 70.5 |
| MIRACL rerank Arabic | 12.0 | 85.5 |
| MIRACL pairwise Arabic | 50.0 | 87.5 |
Model pages: Classify-1 · Tev1 4B Experimental. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.