system-one-bench

~/system-one-bench / compare/celeris-1-decision-vs-together-ai-tev1-4b-experimental

Head to head

Celeris-1 Decision vs Tev1 4B Experimental

Celeris-1 Decision scores 83.6 and Tev1 4B Experimental 71.5 on the System One Index. Celeris-1 Decision is more accurate on 13 of 16 text datasets, Tev1 4B Experimental on 3. Same cases, same request, same scorer.

Overall

Celeris-1 DecisionTev1 4B Experimental
System One Index83.671.5
English83.266.4
Arabic83.175.8
Vision Indexโ€“โ€“
Calibration error2.6%5.3%
Median latencyโ€“232 ms
Cost per 1,000 calls$0.014$0.019

Text datasets, accuracy

Celeris-1 DecisionTev1 4B Experimental
AG News English90.591.5
SST-2 English95.594.0
SST-5 English61.043.0
DAIR Emotion English72.556.5
Banking77 English89.01.0
BoolQ English89.591.0
XNLI English English91.387.3
Typed decisions English76.467.0
XNLI French French90.082.7
XNLI Arabic Arabic80.076.0
MASSIVE scenario Arabic73.556.5
MASSIVE intent Arabic88.070.0
OSACT4 offensive Arabic94.584.5
Tweet sentiment Arabic69.070.5
MIRACL rerank Arabic86.585.5
MIRACL pairwise Arabic90.087.5

Model pages: Celeris-1 Decision ยท Tev1 4B Experimental. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.