~/system-one-bench / compare/baseten-elephant-1-vs-nace-ai-drex-v2-0
Head to head
Elephant 1 scores 82.1 and Drex v2.0 78.6 on the System One Index. Elephant 1 is more accurate on 12 of 16 text datasets, Drex v2.0 on 4. Same cases, same request, same scorer.
| Elephant 1 | Drex v2.0 | |
|---|---|---|
| System One Index | 82.1 | 78.6 |
| English | 81.3 | 77.8 |
| Arabic | 82.0 | 79.6 |
| Vision Index | โ | โ |
| Calibration error | 3.7% | 1.5% |
| Median latency | โ | 294 ms |
| Cost per 1,000 calls | $0.013 | $0.015 |
| Elephant 1 | Drex v2.0 | |
|---|---|---|
| AG News English | 89.0 | 88.0 |
| SST-2 English | 94.5 | 94.0 |
| SST-5 English | 58.5 | 56.0 |
| DAIR Emotion English | 54.5 | 55.5 |
| Banking77 English | 94.5 | 80.5 |
| BoolQ English | 91.0 | 92.5 |
| XNLI English English | 94.0 | 87.3 |
| Typed decisions English | 74.2 | 68.7 |
| XNLI French French | 90.0 | 77.3 |
| XNLI Arabic Arabic | 79.3 | 71.3 |
| MASSIVE scenario Arabic | 69.0 | 63.5 |
| MASSIVE intent Arabic | 89.5 | 82.5 |
| OSACT4 offensive Arabic | 91.0 | 94.5 |
| Tweet sentiment Arabic | 66.0 | 69.5 |
| MIRACL rerank Arabic | 88.0 | 87.0 |
| MIRACL pairwise Arabic | 91.0 | 89.0 |
Model pages: Elephant 1 ยท Drex v2.0. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.