Head to head
Clef Omni scores 79.7 and Drex v2.0 78.6 on the System One Index. Clef Omni is more accurate on 8 of 16 text datasets, Drex v2.0 on 8. Same cases, same request, same scorer.
| Clef Omni | Drex v2.0 | |
|---|---|---|
| System One Index | 79.7 | 78.6 |
| English | 78.6 | 77.8 |
| Arabic | 80.5 | 79.6 |
| Vision Index | 72.2 | – |
| Calibration error | 3.4% | 1.5% |
| Median latency | 208 ms | 294 ms |
| Cost per 1,000 calls | $0.074 | $0.015 |
| Clef Omni | Drex v2.0 | |
|---|---|---|
| AG News English | 89.0 | 88.0 |
| SST-2 English | 97.0 | 94.0 |
| SST-5 English | 51.0 | 56.0 |
| DAIR Emotion English | 56.0 | 55.5 |
| Banking77 English | 96.0 | 80.5 |
| BoolQ English | 90.0 | 92.5 |
| XNLI English English | 82.7 | 87.3 |
| Typed decisions English | 67.0 | 68.7 |
| XNLI French French | 82.7 | 77.3 |
| XNLI Arabic Arabic | 73.3 | 71.3 |
| MASSIVE scenario Arabic | 89.0 | 63.5 |
| MASSIVE intent Arabic | 91.5 | 82.5 |
| OSACT4 offensive Arabic | 90.5 | 94.5 |
| Tweet sentiment Arabic | 50.0 | 69.5 |
| MIRACL rerank Arabic | 83.0 | 87.0 |
| MIRACL pairwise Arabic | 86.5 | 89.0 |
| Clef Omni | Drex v2.0 | |
|---|---|---|
| SST-2, as an image | 89.5 | – |
| Arabic sentiment, as an image | 28.5 | – |
| Oxford-IIIT Pet | 88.0 | – |
| VQAv2 yes/no | 87.0 | – |
| CountBench | 68.0 | – |
Model pages: Clef Omni · Drex v2.0. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.