system-one-bench

~/system-one-bench / compare/celeris-1-decision-vs-jared-palmer-kev-4b

Head to head

Celeris-1 Decision vs Kev 4B

Celeris-1 Decision scores 83.6 and Kev 4B 76.8 on the System One Index. Celeris-1 Decision is more accurate on 14 of 16 text datasets, Kev 4B on 1, and 1 are tied. Same cases, same request, same scorer.

Overall

Celeris-1 DecisionKev 4B
System One Index83.676.8
English83.278.3
Arabic83.174.8
Vision Indexโ€“โ€“
Calibration error2.6%5.1%
Median latencyโ€“552 ms
Cost per 1,000 calls$0.014$0.010

Text datasets, accuracy

Celeris-1 DecisionKev 4B
AG News English90.590.5
SST-2 English95.592.5
SST-5 English61.052.0
DAIR Emotion English72.554.5
Banking77 English89.088.0
BoolQ English89.591.5
XNLI English English91.390.7
Typed decisions English76.466.3
XNLI French French90.079.3
XNLI Arabic Arabic80.072.7
MASSIVE scenario Arabic73.557.0
MASSIVE intent Arabic88.076.5
OSACT4 offensive Arabic94.588.5
Tweet sentiment Arabic69.063.0
MIRACL rerank Arabic86.579.5
MIRACL pairwise Arabic90.086.5

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