~/system-one-bench / compare/cloudflare-clef-flash-vs-patrick-18-dexy-decision-9b
Head to head
Clef Flash scores 80.3 and Dexy-Decision-9B 84.7 on the System One Index. Clef Flash is more accurate on 2 of 16 text datasets, Dexy-Decision-9B on 12, and 2 are tied. Same cases, same request, same scorer.
| Clef Flash | Dexy-Decision-9B | |
|---|---|---|
| System One Index | 80.3 | 84.7 |
| English | 80.2 | 82.3 |
| Arabic | 80.5 | 86.8 |
| Vision Index | 78.0 | – |
| Calibration error | 3.5% | 3.5% |
| Median latency | 267 ms | – |
| Cost per 1,000 calls | $0.044 | – |
| Clef Flash | Dexy-Decision-9B | |
|---|---|---|
| AG News English | 90.0 | 91.5 |
| SST-2 English | 95.0 | 95.5 |
| SST-5 English | 57.0 | 60.0 |
| DAIR Emotion English | 57.0 | 54.0 |
| Banking77 English | 98.5 | 93.0 |
| BoolQ English | 88.5 | 89.5 |
| XNLI English English | 84.7 | 94.7 |
| Typed decisions English | 70.5 | 80.0 |
| XNLI French French | 80.0 | 89.3 |
| XNLI Arabic Arabic | 68.7 | 89.3 |
| MASSIVE scenario Arabic | 89.5 | 89.5 |
| MASSIVE intent Arabic | 90.5 | 91.5 |
| OSACT4 offensive Arabic | 89.5 | 91.5 |
| Tweet sentiment Arabic | 47.0 | 67.0 |
| MIRACL rerank Arabic | 88.0 | 88.0 |
| MIRACL pairwise Arabic | 90.0 | 91.0 |
| Clef Flash | Dexy-Decision-9B | |
|---|---|---|
| SST-2, as an image | 95.5 | – |
| Arabic sentiment, as an image | 33.5 | – |
| Oxford-IIIT Pet | 85.0 | – |
| VQAv2 yes/no | 88.0 | – |
| CountBench | 88.0 | – |
Model pages: Clef Flash · Dexy-Decision-9B. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.