Microsoft·released 2023-09-11
Scored on 5 benchmarks. 0% of its scores are independently reproduced — the rest are self-reported or unverified.
Every score, ranked, colored by evidence status.
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| Winogrande | 46.8accuracy (%) | self-reported· optimized | 2023-09-11 |
| HellaSwag | 30.13accuracy (%) | self-reported· optimized | 2023-09-11 |
| ARC AI2 | 25.87accuracy (%) | self-reported· optimized | 2023-09-11 |
| MMLU | 16.8accuracy (%) | self-reported· optimized | 2023-09-11 |
| OpenBookQA | 16.27accuracy (%) | self-reported· optimized | 2023-09-11 |