system-one-bench

~/system-one-bench / compare/baseten-mammoth-1-vs-microsoft-decision-1

Head to head

Mammoth 1 vs Microsoft-Decision-1

Mammoth 1 scores 82.9 and Microsoft-Decision-1 82.0 on the System One Index. Mammoth 1 is more accurate on 10 of 16 text datasets, Microsoft-Decision-1 on 6. Same cases, same request, same scorer.

Overall

Mammoth 1Microsoft-Decision-1
System One Index82.982.0
English82.581.0
Arabic82.883.2
Vision Indexโ€“โ€“
Calibration error4.9%2.8%
Median latencyโ€“240 ms
Cost per 1,000 calls$0.037$0.010

Text datasets, accuracy

Mammoth 1Microsoft-Decision-1
AG News English92.089.0
SST-2 English96.095.5
SST-5 English59.062.5
DAIR Emotion English59.057.0
Banking77 English94.587.0
BoolQ English92.088.5
XNLI English English91.389.3
Typed decisions English76.479.0
XNLI French French86.082.0
XNLI Arabic Arabic78.772.0
MASSIVE scenario Arabic75.083.5
MASSIVE intent Arabic91.089.0
OSACT4 offensive Arabic91.588.5
Tweet sentiment Arabic64.568.5
MIRACL rerank Arabic89.090.0
MIRACL pairwise Arabic90.091.0

Model pages: Mammoth 1 ยท Microsoft-Decision-1. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.