system-one-bench

~/system-one-bench / compare/baseten-elephant-1-vs-jared-palmer-kev-4b

Head to head

Elephant 1 vs Kev 4B

Elephant 1 scores 82.1 and Kev 4B 76.8 on the System One Index. Elephant 1 is more accurate on 13 of 16 text datasets, Kev 4B on 2, and 1 are tied. Same cases, same request, same scorer.

Overall

Elephant 1Kev 4B
System One Index82.176.8
English81.378.3
Arabic82.074.8
Vision Indexโ€“โ€“
Calibration error3.7%5.1%
Median latencyโ€“552 ms
Cost per 1,000 calls$0.013$0.010

Text datasets, accuracy

Elephant 1Kev 4B
AG News English89.090.5
SST-2 English94.592.5
SST-5 English58.552.0
DAIR Emotion English54.554.5
Banking77 English94.588.0
BoolQ English91.091.5
XNLI English English94.090.7
Typed decisions English74.266.3
XNLI French French90.079.3
XNLI Arabic Arabic79.372.7
MASSIVE scenario Arabic69.057.0
MASSIVE intent Arabic89.576.5
OSACT4 offensive Arabic91.088.5
Tweet sentiment Arabic66.063.0
MIRACL rerank Arabic88.079.5
MIRACL pairwise Arabic91.086.5

Model pages: Elephant 1 ยท Kev 4B. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.