~/system-one-bench / compare/baseten-elephant-1-vs-scaledown-classify-1
Head to head
Elephant 1 scores 82.1 and Classify-1 30.6 on the System One Index. Elephant 1 is more accurate on 15 of 16 text datasets, Classify-1 on 1. Same cases, same request, same scorer.
| Elephant 1 | Classify-1 | |
|---|---|---|
| System One Index | 82.1 | 30.6 |
| English | 81.3 | 30.7 |
| Arabic | 82.0 | 30.6 |
| Vision Index | โ | โ |
| Calibration error | 3.7% | 3.2% |
| Median latency | โ | 254 ms |
| Cost per 1,000 calls | $0.013 | $0.018 |
| Elephant 1 | Classify-1 | |
|---|---|---|
| AG News English | 89.0 | 26.5 |
| SST-2 English | 94.5 | 50.0 |
| SST-5 English | 58.5 | 13.0 |
| DAIR Emotion English | 54.5 | 59.0 |
| Banking77 English | 94.5 | 1.0 |
| BoolQ English | 91.0 | 38.0 |
| XNLI English English | 94.0 | 30.7 |
| Typed decisions English | 74.2 | 27.2 |
| XNLI French French | 90.0 | 30.7 |
| XNLI Arabic Arabic | 79.3 | 30.7 |
| MASSIVE scenario Arabic | 69.0 | 3.0 |
| MASSIVE intent Arabic | 89.5 | 3.5 |
| OSACT4 offensive Arabic | 91.0 | 80.0 |
| Tweet sentiment Arabic | 66.0 | 35.0 |
| MIRACL rerank Arabic | 88.0 | 12.0 |
| MIRACL pairwise Arabic | 91.0 | 50.0 |
Model pages: Elephant 1 ยท Classify-1. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.