~/system-one-bench / models/baseten-mammoth-1
Baseten · OpenRouter API
Mammoth 1Mammoth 1 by Baseten scores 82.9 on the System One Index (7th of 39). Calibration error 4.9%.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 92.0 | 2nd of 39 | 92.5 | 5.8% |
| SST-2 English | 96.0 | 5th of 39 | 98.0 | 4.1% |
| SST-5 English | 59.0 | 11th of 39 | 62.5 | 9.7% |
| DAIR Emotion English | 59.0 | 16th of 39 | 91.5 | 17.0% |
| Banking77 English | 94.5 | 4th of 39 | 98.5 | 2.8% |
| BoolQ English | 92.0 | 3rd of 39 | 93.0 | 4.0% |
| XNLI English English | 91.3 | 8th of 39 | 95.3 | 3.1% |
| Typed decisions English | 76.4 | 4th of 39 | 80.0 | 11.0% |
| XNLI French French | 86.0 | 8th of 39 | 91.3 | 4.3% |
| XNLI Arabic Arabic | 78.7 | 11th of 41 | 90.0 | 4.9% |
| MASSIVE scenario Arabic | 75.0 | 8th of 41 | 89.5 | 11.9% |
| MASSIVE intent Arabic | 91.0 | 3rd of 41 | 91.5 | 3.6% |
| OSACT4 offensive Arabic | 91.5 | 13th of 41 | 97.0 | 1.9% |
| Tweet sentiment Arabic | 64.5 | 24th of 41 | 80.0 | 15.1% |
| MIRACL rerank Arabic | 89.0 | 5th of 41 | 90.5 | 6.2% |
| MIRACL pairwise Arabic | 90.0 | 12th of 41 | 91.5 | 4.2% |
baseten/mammoth-127B, built on Qwen3.5-27B by Base Labs, Baseten's research team. MIT weights (baselabs/mammoth-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.