OpenAI ยท OpenRouter API
GPT-6 Luna Decisions by OpenAI scores 76.0 on the System One Index (14th of 24) and 81.6 on the Vision Index (4th of 8). Calibration error 9.8%, median latency 215 ms.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 87.5 | 17th of 24 | 92.5 | 11.2% |
| SST-2 English | 93.5 | 13th of 24 | 97.0 | 4.8% |
| SST-5 English | 61.5 | 3rd of 24 | 62.5 | 16.0% |
| DAIR Emotion English | 58.5 | 9th of 24 | 91.5 | 26.1% |
| Banking77 English | 77.5 | 18th of 24 | 98.5 | 8.7% |
| BoolQ English | 85.5 | 16th of 24 | 92.5 | 8.6% |
| XNLI English English | 66.7 | 22nd of 24 | 93.3 | 20.7% |
| Typed decisions English | 71.0 | 8th of 24 | 79.0 | 8.4% |
| XNLI French French | 62.0 | 22nd of 24 | 90.7 | 21.5% |
| XNLI Arabic Arabic | 54.0 | 23rd of 26 | 89.3 | 29.8% |
| MASSIVE scenario Arabic | 66.0 | 10th of 26 | 89.5 | 16.8% |
| MASSIVE intent Arabic | 84.5 | 9th of 26 | 91.5 | 5.1% |
| OSACT4 offensive Arabic | 93.5 | 3rd of 26 | 95.5 | 2.2% |
| Tweet sentiment Arabic | 80.0 | 1st of 26 | 80.0 | 10.7% |
| MIRACL rerank Arabic | 86.0 | 10th of 26 | 90.5 | 4.5% |
| MIRACL pairwise Arabic | 87.5 | 10th of 26 | 91.0 | 10.7% |
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| SST-2, as an image | 88.5 | 8th of 8 | 96.0 | 5.5% |
| Arabic sentiment, as an image | 67.0 | 1st of 8 | 67.0 | 13.5% |
| Oxford-IIIT Pet | 87.0 | 3rd of 8 | 88.5 | 4.2% |
| VQAv2 yes/no | 80.5 | 8th of 8 | 89.5 | 8.2% |
| CountBench | 85.0 | 7th of 8 | 94.5 | 9.0% |
openai/gpt-6-luna-decisionsRanks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.