~/system-one-bench / compare/baseten-mammoth-1-vs-nace-ai-drex-v1-5
Head to head
Mammoth 1 scores 82.9 and Drex v1.5 77.6 on the System One Index. Mammoth 1 is more accurate on 15 of 16 text datasets, Drex v1.5 on 0, and 1 are tied. Same cases, same request, same scorer.
| Mammoth 1 | Drex v1.5 | |
|---|---|---|
| System One Index | 82.9 | 77.6 |
| English | 82.5 | 77.6 |
| Arabic | 82.8 | 77.0 |
| Vision Index | โ | โ |
| Calibration error | 4.9% | 10.8% |
| Median latency | โ | 198 ms |
| Cost per 1,000 calls | $0.037 | $0.010 |
| Mammoth 1 | Drex v1.5 | |
|---|---|---|
| AG News English | 92.0 | 90.0 |
| SST-2 English | 96.0 | 94.5 |
| SST-5 English | 59.0 | 53.5 |
| DAIR Emotion English | 59.0 | 53.5 |
| Banking77 English | 94.5 | 87.0 |
| BoolQ English | 92.0 | 89.0 |
| XNLI English English | 91.3 | 88.7 |
| Typed decisions English | 76.4 | 64.8 |
| XNLI French French | 86.0 | 82.0 |
| XNLI Arabic Arabic | 78.7 | 74.0 |
| MASSIVE scenario Arabic | 75.0 | 58.5 |
| MASSIVE intent Arabic | 91.0 | 77.5 |
| OSACT4 offensive Arabic | 91.5 | 90.5 |
| Tweet sentiment Arabic | 64.5 | 63.5 |
| MIRACL rerank Arabic | 89.0 | 89.0 |
| MIRACL pairwise Arabic | 90.0 | 86.0 |
Model pages: Mammoth 1 ยท Drex v1.5. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.