~/system-one-bench / compare/baseten-elephant-1-vs-liquid-ai-d1-1
Head to head
Elephant 1 scores 82.1 and d1.1 82.3 on the System One Index. Elephant 1 is more accurate on 6 of 16 text datasets, d1.1 on 9, and 1 are tied. Same cases, same request, same scorer.
| Elephant 1 | d1.1 | |
|---|---|---|
| System One Index | 82.1 | 82.3 |
| English | 81.3 | 81.3 |
| Arabic | 82.0 | 83.3 |
| Vision Index | – | 85.8 |
| Calibration error | 3.7% | 2.1% |
| Median latency | – | – |
| Cost per 1,000 calls | $0.013 | $0.005 |
| Elephant 1 | d1.1 | |
|---|---|---|
| AG News English | 89.0 | 90.5 |
| SST-2 English | 94.5 | 98.0 |
| SST-5 English | 58.5 | 60.0 |
| DAIR Emotion English | 54.5 | 60.5 |
| Banking77 English | 94.5 | 87.0 |
| BoolQ English | 91.0 | 92.0 |
| XNLI English English | 94.0 | 90.0 |
| Typed decisions English | 74.2 | 72.0 |
| XNLI French French | 90.0 | 83.3 |
| XNLI Arabic Arabic | 79.3 | 79.3 |
| MASSIVE scenario Arabic | 69.0 | 73.0 |
| MASSIVE intent Arabic | 89.5 | 88.0 |
| OSACT4 offensive Arabic | 91.0 | 92.5 |
| Tweet sentiment Arabic | 66.0 | 72.5 |
| MIRACL rerank Arabic | 88.0 | 88.5 |
| MIRACL pairwise Arabic | 91.0 | 89.5 |
| Elephant 1 | d1.1 | |
|---|---|---|
| SST-2, as an image | – | 96.5 |
| Arabic sentiment, as an image | – | 61.0 |
| Oxford-IIIT Pet | – | 90.5 |
| VQAv2 yes/no | – | 89.5 |
| CountBench | – | 91.5 |
Model pages: Elephant 1 · d1.1. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.