~/system-one-bench / compare/meragpt-decider-1-vs-perplexity-decider-flash-v1-2
Head to head
Decider 1 scores 61.8 and Decider Flash V1.2 81.9 on the System One Index. Decider 1 is more accurate on 3 of 16 text datasets, Decider Flash V1.2 on 13. Same cases, same request, same scorer.
| Decider 1 | Decider Flash V1.2 | |
|---|---|---|
| System One Index | 61.8 | 81.9 |
| English | 67.4 | 80.0 |
| Arabic | 54.7 | 83.3 |
| Vision Index | – | 77.5 |
| Calibration error | 7.9% | 1.8% |
| Median latency | – | 104 ms |
| Cost per 1,000 calls | $0.012 | $0.005 |
| Decider 1 | Decider Flash V1.2 | |
|---|---|---|
| AG News English | 87.5 | 89.0 |
| SST-2 English | 93.0 | 95.0 |
| SST-5 English | 55.5 | 54.0 |
| DAIR Emotion English | 67.5 | 57.0 |
| Banking77 English | 1.0 | 90.5 |
| BoolQ English | 80.0 | 89.0 |
| XNLI English English | 80.7 | 92.7 |
| Typed decisions English | 74.2 | 73.0 |
| XNLI French French | 67.3 | 88.0 |
| XNLI Arabic Arabic | 67.3 | 85.3 |
| MASSIVE scenario Arabic | 3.0 | 70.5 |
| MASSIVE intent Arabic | 3.5 | 86.0 |
| OSACT4 offensive Arabic | 86.0 | 90.5 |
| Tweet sentiment Arabic | 64.0 | 71.5 |
| MIRACL rerank Arabic | 81.0 | 88.5 |
| MIRACL pairwise Arabic | 78.0 | 90.5 |
| Decider 1 | Decider Flash V1.2 | |
|---|---|---|
| SST-2, as an image | – | 93.5 |
| Arabic sentiment, as an image | – | 35.0 |
| Oxford-IIIT Pet | – | 84.0 |
| VQAv2 yes/no | – | 87.5 |
| CountBench | – | 87.5 |
Model pages: Decider 1 · Decider Flash V1.2. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.