~/system-one-bench / compare/google-gemini-decision-1-vs-scaledown-classify-1
Head to head
Gemini Decision 1 scores 81.6 and Classify-1 30.6 on the System One Index. Gemini Decision 1 is more accurate on 16 of 16 text datasets, Classify-1 on 0. Same cases, same request, same scorer.
| Gemini Decision 1 | Classify-1 | |
|---|---|---|
| System One Index | 81.6 | 30.6 |
| English | 79.2 | 30.7 |
| Arabic | 84.5 | 30.6 |
| Vision Index | 89.2 | – |
| Calibration error | 3.6% | 3.2% |
| Median latency | – | 254 ms |
| Cost per 1,000 calls | $0.036 | $0.018 |
| Gemini Decision 1 | Classify-1 | |
|---|---|---|
| AG News English | 88.5 | 26.5 |
| SST-2 English | 96.5 | 50.0 |
| SST-5 English | 59.0 | 13.0 |
| DAIR Emotion English | 59.5 | 59.0 |
| Banking77 English | 79.5 | 1.0 |
| BoolQ English | 89.0 | 38.0 |
| XNLI English English | 87.3 | 30.7 |
| Typed decisions English | 74.0 | 27.2 |
| XNLI French French | 81.3 | 30.7 |
| XNLI Arabic Arabic | 78.0 | 30.7 |
| MASSIVE scenario Arabic | 72.5 | 3.0 |
| MASSIVE intent Arabic | 87.0 | 3.5 |
| OSACT4 offensive Arabic | 95.5 | 80.0 |
| Tweet sentiment Arabic | 78.0 | 35.0 |
| MIRACL rerank Arabic | 89.5 | 12.0 |
| MIRACL pairwise Arabic | 91.0 | 50.0 |
| Gemini Decision 1 | Classify-1 | |
|---|---|---|
| SST-2, as an image | 94.5 | – |
| Arabic sentiment, as an image | 75.0 | – |
| Oxford-IIIT Pet | 98.0 | – |
| VQAv2 yes/no | 86.0 | – |
| CountBench | 92.5 | – |
Model pages: Gemini Decision 1 · Classify-1. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.