~/system-one-bench / compare/cloudflare-clef-omni-vs-google-gemini-decision-1
Head to head
Clef Omni scores 79.7 and Gemini Decision 1 81.6 on the System One Index. Clef Omni is more accurate on 7 of 16 text datasets, Gemini Decision 1 on 9. Same cases, same request, same scorer.
| Clef Omni | Gemini Decision 1 | |
|---|---|---|
| System One Index | 79.7 | 81.6 |
| English | 78.6 | 79.2 |
| Arabic | 80.5 | 84.5 |
| Vision Index | 72.2 | 89.2 |
| Calibration error | 3.4% | 3.6% |
| Median latency | 208 ms | – |
| Cost per 1,000 calls | $0.074 | $0.036 |
| Clef Omni | Gemini Decision 1 | |
|---|---|---|
| AG News English | 89.0 | 88.5 |
| SST-2 English | 97.0 | 96.5 |
| SST-5 English | 51.0 | 59.0 |
| DAIR Emotion English | 56.0 | 59.5 |
| Banking77 English | 96.0 | 79.5 |
| BoolQ English | 90.0 | 89.0 |
| XNLI English English | 82.7 | 87.3 |
| Typed decisions English | 67.0 | 74.0 |
| XNLI French French | 82.7 | 81.3 |
| XNLI Arabic Arabic | 73.3 | 78.0 |
| MASSIVE scenario Arabic | 89.0 | 72.5 |
| MASSIVE intent Arabic | 91.5 | 87.0 |
| OSACT4 offensive Arabic | 90.5 | 95.5 |
| Tweet sentiment Arabic | 50.0 | 78.0 |
| MIRACL rerank Arabic | 83.0 | 89.5 |
| MIRACL pairwise Arabic | 86.5 | 91.0 |
| Clef Omni | Gemini Decision 1 | |
|---|---|---|
| SST-2, as an image | 89.5 | 94.5 |
| Arabic sentiment, as an image | 28.5 | 75.0 |
| Oxford-IIIT Pet | 88.0 | 98.0 |
| VQAv2 yes/no | 87.0 | 86.0 |
| CountBench | 68.0 | 92.5 |
Model pages: Clef Omni · Gemini Decision 1. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.