HardcoreLogic - Unsolvable Puzzles: leaderboard
The Unsolvable Puzzles slice of HardcoreLogic — recognizing inconsistent logic-puzzle constraints instead of inventing a solution; accuracy.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 98.52 | #186 |
| 2 | GPT-5 | 97.78 | #98 |
| 3 | Grok 4 | 97.59 | #181 |
| 4 | DeepSeek R1 0528 Qwen3 8B | 95.19 | #764 |
| 5 | O4 Mini | 95 | #180 |
| 6 | Grok 3 Mini | 94.63 | #281 |
| 7 | DeepSeek R1 0528 | 93.5 | #222 |
| 8 | GPT-OSS-120B | 93.35 | #341 |
| 9 | GLM-4.5 | 93.26 | #275 |
| 10 | Gemini 2.5 Pro | 91.48 | #152 |
| 11 | DeepSeek V3.1 | 88.76 | #264 |
| 12 | Kimi K2 | 87.46 | #243 |
| 13 | Qwen 3 30B A3B 2507 (Thinking) | 86.09 | #376 (Qwen 3 30B A3B 2507) |
| 14 | Seed-OSS-36B-Instruct | 85.76 | #366 |
| 15 | Qwen 3 235B A22B 2507 (Thinking) | 84.41 | #260 (Qwen 3 235B A22B 2507) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=hardcorelogic-unsolvable-puzzles · How It Works · Data refreshed daily, snapshot 2026-10-20.