card: model-index + GPQA 88.38% badge + .eval_results auto-leaderboard + abstract proprietary cascade
f823c06 verified | # Darwin-60B-DUO Benchmarks | |
| > Updated 2026-05-28 KST | |
| ## 🏆 GPQA Diamond (full 198 questions) — **FINAL: 88.38%** | |
| | Stage | Score | | |
| |-------|-------| | |
| | Darwin single (MAJ@5) | 69.70% | | |
| | AWAXIS single (MAJ@5) | 77.27% | | |
| | Cascade (+ MTI) — first cascade | 83.84% | | |
| | **🏆 Darwin-DELPHI DUO FINAL** | **88.38%** | | |
| The final score is produced by the **Darwin-DELPHI test-time engine** operating over the two-specialist DUO. The engine is proprietary. | |
| ## 📌 Constituent verified scores | |
| | Constituent | Benchmark | Rank | | |
| |-------------|-----------|------| | |
| | Darwin-28B-REASON | Hugging Face GPQA Diamond | #3 | | |
| | AWAXIS-Think-31B | National K-AI Leaderboard (Republic of Korea, MSIT) | #1 | | |
| ## Future evaluations (scheduled) | |
| | Benchmark | Status | | |
| |-----------|--------| | |
| | KMMLU | TBA | | |
| | CLIcK | TBA | | |
| | Helmet / Ruler (long context) | TBA | | |
| | Needle-in-Haystack 128 K | NIAH 32 K verified per backend | | |