~/system-one-bench / models/baseten-elephant-1
Baseten · OpenRouter API
Elephant 1Elephant 1 by Baseten scores 82.1 on the System One Index (11th of 39). Calibration error 3.7%.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 89.0 | 18th of 39 | 92.5 | 5.1% |
| SST-2 English | 94.5 | 20th of 39 | 98.0 | 2.9% |
| SST-5 English | 58.5 | 14th of 39 | 62.5 | 12.9% |
| DAIR Emotion English | 54.5 | 32nd of 39 | 91.5 | 22.0% |
| Banking77 English | 94.5 | 4th of 39 | 98.5 | 2.9% |
| BoolQ English | 91.0 | 9th of 39 | 93.0 | 4.4% |
| XNLI English English | 94.0 | 4th of 39 | 95.3 | 4.0% |
| Typed decisions English | 74.2 | 9th of 39 | 80.0 | 10.5% |
| XNLI French French | 90.0 | 2nd of 39 | 91.3 | 5.4% |
| XNLI Arabic Arabic | 79.3 | 9th of 41 | 90.0 | 7.4% |
| MASSIVE scenario Arabic | 69.0 | 21st of 41 | 89.5 | 14.5% |
| MASSIVE intent Arabic | 89.5 | 8th of 41 | 91.5 | 3.9% |
| OSACT4 offensive Arabic | 91.0 | 15th of 41 | 97.0 | 3.5% |
| Tweet sentiment Arabic | 66.0 | 21st of 41 | 80.0 | 13.7% |
| MIRACL rerank Arabic | 88.0 | 12th of 41 | 90.5 | 5.0% |
| MIRACL pairwise Arabic | 91.0 | 3rd of 41 | 91.5 | 1.6% |
baseten/elephant-19B, built on Qwen3.5-9B by Base Labs, Baseten's research team. MIT weights (baselabs/elephant-1). Text only.
Ranks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.