~/system-one-bench / compare/celeris-1-decision-vs-respan-span-02
Head to head
Celeris-1 Decision scores 83.6 and Span-02 79.5 on the System One Index. Celeris-1 Decision is more accurate on 10 of 16 text datasets, Span-02 on 5, and 1 are tied. Same cases, same request, same scorer.
| Celeris-1 Decision | Span-02 | |
|---|---|---|
| System One Index | 83.6 | 79.5 |
| English | 83.2 | 77.9 |
| Arabic | 83.1 | 82.3 |
| Vision Index | – | 86.1 |
| Calibration error | 2.6% | 10.5% |
| Median latency | – | 204 ms |
| Cost per 1,000 calls | $0.014 | $0.005 |
| Celeris-1 Decision | Span-02 | |
|---|---|---|
| AG News English | 90.5 | 86.5 |
| SST-2 English | 95.5 | 95.5 |
| SST-5 English | 61.0 | 57.5 |
| DAIR Emotion English | 72.5 | 56.0 |
| Banking77 English | 89.0 | 82.0 |
| BoolQ English | 89.5 | 90.0 |
| XNLI English English | 91.3 | 83.3 |
| Typed decisions English | 76.4 | 72.1 |
| XNLI French French | 90.0 | 73.3 |
| XNLI Arabic Arabic | 80.0 | 66.0 |
| MASSIVE scenario Arabic | 73.5 | 71.0 |
| MASSIVE intent Arabic | 88.0 | 86.0 |
| OSACT4 offensive Arabic | 94.5 | 97.0 |
| Tweet sentiment Arabic | 69.0 | 76.0 |
| MIRACL rerank Arabic | 86.5 | 88.5 |
| MIRACL pairwise Arabic | 90.0 | 91.5 |
| Celeris-1 Decision | Span-02 | |
|---|---|---|
| SST-2, as an image | – | 92.5 |
| Arabic sentiment, as an image | – | 70.0 |
| Oxford-IIIT Pet | – | 92.5 |
| VQAv2 yes/no | – | 87.0 |
| CountBench | – | 88.5 |
Model pages: Celeris-1 Decision · Span-02. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.