system-one-bench

~/system-one-bench / compare/liquid-ai-d1-1-vs-vllm-semantic-router-d3-nano

Head to head

d1.1 vs d3 nano

d1.1 scores 82.3 and d3 nano 78.0 on the System One Index. d1.1 is more accurate on 13 of 16 text datasets, d3 nano on 3. Same cases, same request, same scorer.

Overall

d1.1d3 nano
System One Index82.378.0
English81.377.9
Arabic83.376.9
Vision Index85.8–
Calibration error2.1%7.0%
Median latency––
Cost per 1,000 calls$0.005–

Text datasets, accuracy

d1.1d3 nano
AG News English90.588.0
SST-2 English98.089.5
SST-5 English60.055.5
DAIR Emotion English60.559.5
Banking77 English87.093.0
BoolQ English92.088.0
XNLI English English90.087.3
Typed decisions English72.062.6
XNLI French French83.386.0
XNLI Arabic Arabic79.382.0
MASSIVE scenario Arabic73.061.5
MASSIVE intent Arabic88.081.0
OSACT4 offensive Arabic92.579.0
Tweet sentiment Arabic72.571.0
MIRACL rerank Arabic88.577.5
MIRACL pairwise Arabic89.586.5

Image datasets, accuracy

d1.1d3 nano
SST-2, as an image96.5–
Arabic sentiment, as an image61.0–
Oxford-IIIT Pet90.5–
VQAv2 yes/no89.5–
CountBench91.5–

Model pages: d1.1 · d3 nano. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.