system-one-bench

~/system-one-bench / compare/cloudflare-clef-vs-meragpt-decider-1

Head to head

Clef vs Decider 1

Clef scores 81.6 and Decider 1 61.8 on the System One Index. Clef is more accurate on 12 of 16 text datasets, Decider 1 on 4. Same cases, same request, same scorer.

Overall

ClefDecider 1
System One Index81.661.8
English80.667.4
Arabic83.054.7
Vision Index78.1–
Calibration error2.2%7.9%
Median latency310 ms–
Cost per 1,000 calls$0.116$0.012

Text datasets, accuracy

ClefDecider 1
AG News English90.587.5
SST-2 English97.093.0
SST-5 English51.555.5
DAIR Emotion English58.567.5
Banking77 English95.51.0
BoolQ English91.080.0
XNLI English English88.780.7
Typed decisions English72.274.2
XNLI French French80.067.3
XNLI Arabic Arabic78.067.3
MASSIVE scenario Arabic88.03.0
MASSIVE intent Arabic89.53.5
OSACT4 offensive Arabic89.586.0
Tweet sentiment Arabic58.064.0
MIRACL rerank Arabic89.081.0
MIRACL pairwise Arabic89.078.0

Image datasets, accuracy

ClefDecider 1
SST-2, as an image95.5–
Arabic sentiment, as an image35.0–
Oxford-IIIT Pet82.5–
VQAv2 yes/no87.5–
CountBench90.0–

Model pages: Clef · Decider 1. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.