system-one-bench

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

Head to head

Mammoth 1 vs GPT-6 Luna Decisions

Mammoth 1 scores 82.9 and GPT-6 Luna Decisions 76.0 on the System One Index. Mammoth 1 is more accurate on 13 of 16 text datasets, GPT-6 Luna Decisions on 3. Same cases, same request, same scorer.

Overall

Mammoth 1GPT-6 Luna Decisions
System One Index82.976.0
English82.575.2
Arabic82.878.8
Vision Index–81.6
Calibration error4.9%9.8%
Median latency–215 ms
Cost per 1,000 calls$0.037$0.040

Text datasets, accuracy

Mammoth 1GPT-6 Luna Decisions
AG News English92.087.5
SST-2 English96.093.5
SST-5 English59.061.5
DAIR Emotion English59.058.5
Banking77 English94.577.5
BoolQ English92.085.5
XNLI English English91.366.7
Typed decisions English76.471.0
XNLI French French86.062.0
XNLI Arabic Arabic78.754.0
MASSIVE scenario Arabic75.066.0
MASSIVE intent Arabic91.084.5
OSACT4 offensive Arabic91.593.5
Tweet sentiment Arabic64.580.0
MIRACL rerank Arabic89.086.0
MIRACL pairwise Arabic90.087.5

Image datasets, accuracy

Mammoth 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: Mammoth 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.