system-one-bench

~/system-one-bench / compare/fastino-glide-vs-liquid-ai-d1-ultra

Head to head

GLiDE vs d1-ultra

GLiDE scores 84.1 and d1-ultra 82.9 on the System One Index. GLiDE is more accurate on 10 of 16 text datasets, d1-ultra on 5, and 1 are tied. Same cases, same request, same scorer.

Overall

GLiDEd1-ultra
System One Index84.182.9
English83.081.7
Arabic84.884.2
Vision Index82.585.6
Calibration error3.6%1.0%
Median latency–168 ms
Cost per 1,000 calls$0.017$0.010

Text datasets, accuracy

GLiDEd1-ultra
AG News English92.091.0
SST-2 English96.095.5
SST-5 English61.561.5
DAIR Emotion English61.563.0
Banking77 English91.583.5
BoolQ English92.091.5
XNLI English English94.792.7
Typed decisions English75.075.0
XNLI French French88.083.3
XNLI Arabic Arabic83.376.0
MASSIVE scenario Arabic73.576.0
MASSIVE intent Arabic91.087.0
OSACT4 offensive Arabic95.094.5
Tweet sentiment Arabic73.576.0
MIRACL rerank Arabic87.089.0
MIRACL pairwise Arabic90.591.0

Image datasets, accuracy

GLiDEd1-ultra
SST-2, as an image94.097.0
Arabic sentiment, as an image51.562.5
Oxford-IIIT Pet84.587.5
VQAv2 yes/no89.089.5
CountBench93.591.5

Model pages: GLiDE · d1-ultra. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.