system-one-bench

~/system-one-bench / compare/baseten-joey-1-vs-nace-ai-drex-v1-5

Head to head

Joey 1 vs Drex v1.5

Joey 1 scores 73.3 and Drex v1.5 77.6 on the System One Index. Joey 1 is more accurate on 4 of 16 text datasets, Drex v1.5 on 12. Same cases, same request, same scorer.

Overall

Joey 1Drex v1.5
System One Index73.377.6
English72.977.6
Arabic73.577.0
Vision Indexโ€“โ€“
Calibration error2.1%10.8%
Median latencyโ€“198 ms
Cost per 1,000 calls$0.004$0.010

Text datasets, accuracy

Joey 1Drex v1.5
AG News English85.090.0
SST-2 English86.094.5
SST-5 English40.553.5
DAIR Emotion English52.553.5
Banking77 English93.087.0
BoolQ English80.089.0
XNLI English English84.088.7
Typed decisions English62.464.8
XNLI French French74.782.0
XNLI Arabic Arabic64.774.0
MASSIVE scenario Arabic71.558.5
MASSIVE intent Arabic84.577.5
OSACT4 offensive Arabic83.090.5
Tweet sentiment Arabic66.063.5
MIRACL rerank Arabic62.089.0
MIRACL pairwise Arabic83.086.0

Model pages: Joey 1 ยท Drex v1.5. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.