system-one-bench

~/system-one-bench / compare/celeris-1-decision-vs-nace-ai-drex-v2-0

Head to head

Celeris-1 Decision vs Drex v2.0

Celeris-1 Decision scores 83.6 and Drex v2.0 78.6 on the System One Index. Celeris-1 Decision is more accurate on 12 of 16 text datasets, Drex v2.0 on 3, and 1 are tied. Same cases, same request, same scorer.

Overall

Celeris-1 DecisionDrex v2.0
System One Index83.678.6
English83.277.8
Arabic83.179.6
Vision Indexโ€“โ€“
Calibration error2.6%1.5%
Median latencyโ€“294 ms
Cost per 1,000 calls$0.014$0.015

Text datasets, accuracy

Celeris-1 DecisionDrex v2.0
AG News English90.588.0
SST-2 English95.594.0
SST-5 English61.056.0
DAIR Emotion English72.555.5
Banking77 English89.080.5
BoolQ English89.592.5
XNLI English English91.387.3
Typed decisions English76.468.7
XNLI French French90.077.3
XNLI Arabic Arabic80.071.3
MASSIVE scenario Arabic73.563.5
MASSIVE intent Arabic88.082.5
OSACT4 offensive Arabic94.594.5
Tweet sentiment Arabic69.069.5
MIRACL rerank Arabic86.587.0
MIRACL pairwise Arabic90.089.0

Model pages: Celeris-1 Decision ยท Drex v2.0. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.