System One Bench

Head to head

Mercury Decide vs GPT-6 Luna Decisions

Mercury Decide scores 82.4 and GPT-6 Luna Decisions 76.0 on the System One Index. Mercury Decide is more accurate on 13 of 16 text datasets, GPT-6 Luna Decisions on 3. Same cases, same request, same scorer.

Overall

Mercury DecideGPT-6 Luna Decisions
System One Index82.476.0
English83.775.2
Arabic80.878.8
Vision Index79.481.6
Calibration error13.1%9.8%
Median latency276 ms215 ms
Cost per 1,000 callsFree$0.040

Text datasets, accuracy

Mercury DecideGPT-6 Luna Decisions
AG News English92.087.5
SST-2 English95.593.5
SST-5 English52.061.5
DAIR Emotion English91.558.5
Banking77 English86.077.5
BoolQ English90.585.5
XNLI English English91.366.7
Typed decisions English71.171.0
XNLI French French82.762.0
XNLI Arabic Arabic76.054.0
MASSIVE scenario Arabic72.066.0
MASSIVE intent Arabic85.084.5
OSACT4 offensive Arabic92.093.5
Tweet sentiment Arabic62.580.0
MIRACL rerank Arabic88.086.0
MIRACL pairwise Arabic90.087.5

Image datasets, accuracy

Mercury DecideGPT-6 Luna Decisions
SST-2, as an image93.088.5
Arabic sentiment, as an image39.067.0
Oxford-IIIT Pet84.587.0
VQAv2 yes/no87.080.5
CountBench93.585.0

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