~/system-one-bench / compare/baseten-mammoth-1-vs-liquid-ai-d1
Head to head
Mammoth 1 scores 82.9 and d1 82.7 on the System One Index. Mammoth 1 is more accurate on 9 of 16 text datasets, d1 on 6, and 1 are tied. Same cases, same request, same scorer.
| Mammoth 1 | d1 | |
|---|---|---|
| System One Index | 82.9 | 82.7 |
| English | 82.5 | 81.3 |
| Arabic | 82.8 | 83.9 |
| Vision Index | โ | โ |
| Calibration error | 4.9% | 2.9% |
| Median latency | โ | 188 ms |
| Cost per 1,000 calls | $0.037 | $0.015 |
| Mammoth 1 | d1 | |
|---|---|---|
| AG News English | 92.0 | 90.0 |
| SST-2 English | 96.0 | 95.5 |
| SST-5 English | 59.0 | 56.5 |
| DAIR Emotion English | 59.0 | 63.5 |
| Banking77 English | 94.5 | 89.0 |
| BoolQ English | 92.0 | 90.0 |
| XNLI English English | 91.3 | 91.3 |
| Typed decisions English | 76.4 | 74.5 |
| XNLI French French | 86.0 | 84.7 |
| XNLI Arabic Arabic | 78.7 | 80.0 |
| MASSIVE scenario Arabic | 75.0 | 73.0 |
| MASSIVE intent Arabic | 91.0 | 90.0 |
| OSACT4 offensive Arabic | 91.5 | 95.5 |
| Tweet sentiment Arabic | 64.5 | 67.5 |
| MIRACL rerank Arabic | 89.0 | 90.5 |
| MIRACL pairwise Arabic | 90.0 | 91.0 |
Model pages: Mammoth 1 ยท d1. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.