System One Bench

Head to head

Clef Flash vs GPT-6 Luna Decisions

Clef Flash scores 80.3 and GPT-6 Luna Decisions 76.0 on the System One Index. Clef Flash is more accurate on 11 of 16 text datasets, GPT-6 Luna Decisions on 5. Same cases, same request, same scorer.

Overall

Clef FlashGPT-6 Luna Decisions
System One Index80.376.0
English80.275.2
Arabic80.578.8
Vision Index78.081.6
Calibration error3.5%9.8%
Median latency267 ms215 ms
Cost per 1,000 calls$0.044$0.040

Text datasets, accuracy

Clef FlashGPT-6 Luna Decisions
AG News English90.087.5
SST-2 English95.093.5
SST-5 English57.061.5
DAIR Emotion English57.058.5
Banking77 English98.577.5
BoolQ English88.585.5
XNLI English English84.766.7
Typed decisions English70.571.0
XNLI French French80.062.0
XNLI Arabic Arabic68.754.0
MASSIVE scenario Arabic89.566.0
MASSIVE intent Arabic90.584.5
OSACT4 offensive Arabic89.593.5
Tweet sentiment Arabic47.080.0
MIRACL rerank Arabic88.086.0
MIRACL pairwise Arabic90.087.5

Image datasets, accuracy

Clef FlashGPT-6 Luna Decisions
SST-2, as an image95.588.5
Arabic sentiment, as an image33.567.0
Oxford-IIIT Pet85.087.0
VQAv2 yes/no88.080.5
CountBench88.085.0

Model pages: Clef Flash ยท GPT-6 Luna Decisions. Latency for API models is the median per call through OpenRouter, timed from one machine.