Paydirt - Held Pair Accuracy: leaderboard
Metric: Correct pairs (%); both answers must pass, raw arm, omp harness, 18 held pairs. Source: bountyoperator.com. Saturation forecast: Around January 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 94.44 | #6 (GPT-6 Astra) |
| 2 | GPT-6.1 Sol (Max) | 94.44 | #9 (GPT-6.1 Sol) |
| 3 | Kimi K3 (Max) | 88.89 | #27 (Kimi K3) |
| 4 | GPT-5.6 Terra (Max) | 88.89 | #40 (GPT-5.6 Terra) |
| 5 | GPT-6 Luna (Max) | 83.33 | #51 (GPT-6 Luna) |
| 6 | Claude Opus 5.5 (Max) | 83.33 | #3 (Claude Opus 5.5) |
| 7 | GLM-5.3 (Max) | 77.78 | #38 (GLM-5.3) |
| 8 | DeepSeek V4.1 Flash (Max) | 77.78 | #35 (DeepSeek V4.1 Flash) |
| 9 | DeepSeek V4 Pro (0813) (Max) | 77.78 | #50 (DeepSeek V4 Pro (0813)) |
| 10 | Claude Fable 5.1 (Max) | 77.78 | #8 (Claude Fable 5.1) |
| 11 | Hy4 Preview (High) | 77.78 | |
| 12 | GLM-5.3 Flash (Max) | 72.22 | #62 (GLM-5.3 Flash) |
| 13 | MiniMax M3 (High) | 61.11 | |
| 14 | Gemini 3.8 Flash (High) | 55.56 | #20 (Gemini 3.8 Flash) |
| 15 | Grok 4.7 (xHigh) | 55.56 | #25 (Grok 4.7) |
Interactive version: theaggregate.ai/benchmark?slug=paydirt-held-pair-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-13.