system-one-bench

~/system-one-bench / compare/google-gemini-decision-1-vs-openai-gpt-6-luna-decisions

Head to head

Gemini Decision 1 vs GPT-6 Luna Decisions

Gemini Decision 1 scores 81.6 and GPT-6 Luna Decisions 76.0 on the System One Index. Gemini Decision 1 is more accurate on 14 of 16 text datasets, GPT-6 Luna Decisions on 2. Same cases, same request, same scorer.

Overall

Gemini Decision 1GPT-6 Luna Decisions
System One Index81.676.0
English79.275.2
Arabic84.578.8
Vision Index89.281.6
Calibration error3.6%9.8%
Median latency–215 ms
Cost per 1,000 calls$0.036$0.040

Text datasets, accuracy

Gemini Decision 1GPT-6 Luna Decisions
AG News English88.587.5
SST-2 English96.593.5
SST-5 English59.061.5
DAIR Emotion English59.558.5
Banking77 English79.577.5
BoolQ English89.085.5
XNLI English English87.366.7
Typed decisions English74.071.0
XNLI French French81.362.0
XNLI Arabic Arabic78.054.0
MASSIVE scenario Arabic72.566.0
MASSIVE intent Arabic87.084.5
OSACT4 offensive Arabic95.593.5
Tweet sentiment Arabic78.080.0
MIRACL rerank Arabic89.586.0
MIRACL pairwise Arabic91.087.5

Image datasets, accuracy

Gemini Decision 1GPT-6 Luna Decisions
SST-2, as an image94.588.5
Arabic sentiment, as an image75.067.0
Oxford-IIIT Pet98.087.0
VQAv2 yes/no86.080.5
CountBench92.585.0

Model pages: Gemini Decision 1 · GPT-6 Luna Decisions. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.