~/system-one-bench / compare/baseten-elephant-1-vs-baseten-mammoth-1
Head to head
Elephant 1 scores 82.1 and Mammoth 1 82.9 on the System One Index. Elephant 1 is more accurate on 5 of 16 text datasets, Mammoth 1 on 10, and 1 are tied. Same cases, same request, same scorer.
| Elephant 1 | Mammoth 1 | |
|---|---|---|
| System One Index | 82.1 | 82.9 |
| English | 81.3 | 82.5 |
| Arabic | 82.0 | 82.8 |
| Vision Index | – | – |
| Calibration error | 3.7% | 4.9% |
| Median latency | – | – |
| Cost per 1,000 calls | $0.013 | $0.037 |
| Elephant 1 | Mammoth 1 | |
|---|---|---|
| AG News English | 89.0 | 92.0 |
| SST-2 English | 94.5 | 96.0 |
| SST-5 English | 58.5 | 59.0 |
| DAIR Emotion English | 54.5 | 59.0 |
| Banking77 English | 94.5 | 94.5 |
| BoolQ English | 91.0 | 92.0 |
| XNLI English English | 94.0 | 91.3 |
| Typed decisions English | 74.2 | 76.4 |
| XNLI French French | 90.0 | 86.0 |
| XNLI Arabic Arabic | 79.3 | 78.7 |
| MASSIVE scenario Arabic | 69.0 | 75.0 |
| MASSIVE intent Arabic | 89.5 | 91.0 |
| OSACT4 offensive Arabic | 91.0 | 91.5 |
| Tweet sentiment Arabic | 66.0 | 64.5 |
| MIRACL rerank Arabic | 88.0 | 89.0 |
| MIRACL pairwise Arabic | 91.0 | 90.0 |
Model pages: Elephant 1 · Mammoth 1. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.