Head to head
GPT-6 Luna Decisions scores 76.0 and Span-02 79.5 on the System One Index. GPT-6 Luna Decisions is more accurate on 4 of 16 text datasets, Span-02 on 12. Same cases, same request, same scorer.
| GPT-6 Luna Decisions | Span-02 | |
|---|---|---|
| System One Index | 76.0 | 79.5 |
| English | 75.2 | 77.9 |
| Arabic | 78.8 | 82.3 |
| Vision Index | 81.6 | 86.1 |
| Calibration error | 9.8% | 10.5% |
| Median latency | 215 ms | 204 ms |
| Cost per 1,000 calls | $0.040 | $0.005 |
| GPT-6 Luna Decisions | Span-02 | |
|---|---|---|
| AG News English | 87.5 | 86.5 |
| SST-2 English | 93.5 | 95.5 |
| SST-5 English | 61.5 | 57.5 |
| DAIR Emotion English | 58.5 | 56.0 |
| Banking77 English | 77.5 | 82.0 |
| BoolQ English | 85.5 | 90.0 |
| XNLI English English | 66.7 | 83.3 |
| Typed decisions English | 71.0 | 72.1 |
| XNLI French French | 62.0 | 73.3 |
| XNLI Arabic Arabic | 54.0 | 66.0 |
| MASSIVE scenario Arabic | 66.0 | 71.0 |
| MASSIVE intent Arabic | 84.5 | 86.0 |
| OSACT4 offensive Arabic | 93.5 | 97.0 |
| Tweet sentiment Arabic | 80.0 | 76.0 |
| MIRACL rerank Arabic | 86.0 | 88.5 |
| MIRACL pairwise Arabic | 87.5 | 91.5 |
| GPT-6 Luna Decisions | Span-02 | |
|---|---|---|
| SST-2, as an image | 88.5 | 92.5 |
| Arabic sentiment, as an image | 67.0 | 70.0 |
| Oxford-IIIT Pet | 87.0 | 92.5 |
| VQAv2 yes/no | 80.5 | 87.0 |
| CountBench | 85.0 | 88.5 |
Model pages: GPT-6 Luna Decisions ยท Span-02. Latency for API models is the median per call through OpenRouter, timed from a cloud server with 8 calls in flight.