system-one-bench

~/system-one-bench / compare/baseten-elephant-1-vs-openai-gpt-6-luna-decisions

Head to head

Elephant 1 vs GPT-6 Luna Decisions

Elephant 1 scores 82.1 and GPT-6 Luna Decisions 76.0 on the System One Index. Elephant 1 is more accurate on 12 of 16 text datasets, GPT-6 Luna Decisions on 4. Same cases, same request, same scorer.

Overall

Elephant 1GPT-6 Luna Decisions
System One Index82.176.0
English81.375.2
Arabic82.078.8
Vision Index–81.6
Calibration error3.7%9.8%
Median latency–215 ms
Cost per 1,000 calls$0.013$0.040

Text datasets, accuracy

Elephant 1GPT-6 Luna Decisions
AG News English89.087.5
SST-2 English94.593.5
SST-5 English58.561.5
DAIR Emotion English54.558.5
Banking77 English94.577.5
BoolQ English91.085.5
XNLI English English94.066.7
Typed decisions English74.271.0
XNLI French French90.062.0
XNLI Arabic Arabic79.354.0
MASSIVE scenario Arabic69.066.0
MASSIVE intent Arabic89.584.5
OSACT4 offensive Arabic91.093.5
Tweet sentiment Arabic66.080.0
MIRACL rerank Arabic88.086.0
MIRACL pairwise Arabic91.087.5

Image datasets, accuracy

Elephant 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: Elephant 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.