system-one-bench

~/system-one-bench / compare/meragpt-decider-1-vs-openai-gpt-6-luna-decisions

Head to head

Decider 1 vs GPT-6 Luna Decisions

Decider 1 scores 61.8 and GPT-6 Luna Decisions 76.0 on the System One Index. Decider 1 is more accurate on 5 of 16 text datasets, GPT-6 Luna Decisions on 10, and 1 are tied. Same cases, same request, same scorer.

Overall

Decider 1GPT-6 Luna Decisions
System One Index61.876.0
English67.475.2
Arabic54.778.8
Vision Index–81.6
Calibration error7.9%9.8%
Median latency–215 ms
Cost per 1,000 calls$0.012$0.040

Text datasets, accuracy

Decider 1GPT-6 Luna Decisions
AG News English87.587.5
SST-2 English93.093.5
SST-5 English55.561.5
DAIR Emotion English67.558.5
Banking77 English1.077.5
BoolQ English80.085.5
XNLI English English80.766.7
Typed decisions English74.271.0
XNLI French French67.362.0
XNLI Arabic Arabic67.354.0
MASSIVE scenario Arabic3.066.0
MASSIVE intent Arabic3.584.5
OSACT4 offensive Arabic86.093.5
Tweet sentiment Arabic64.080.0
MIRACL rerank Arabic81.086.0
MIRACL pairwise Arabic78.087.5

Image datasets, accuracy

Decider 1GPT-6 Luna Decisions
SST-2, as an image–88.5
Arabic sentiment, as an image–67.0
Oxford-IIIT Pet–87.0
VQAv2 yes/no–80.5
CountBench–85.0

Model pages: Decider 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.