HardcoreLogic - Total: leaderboard
HardcoreLogic overall — aggregate performance on adversarially difficult logic puzzles, including uncommon, more complex and unsatisfiable cases; solve rate.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around January 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 69.1 | #98 |
| 2 | Grok 4 | 59.55 | #181 |
| 3 | GPT-5 Mini | 54.49 | #186 |
| 4 | GPT-OSS-120B | 51.97 | #341 |
| 5 | O4 Mini | 50.13 | #180 |
| 6 | Qwen 3 235B A22B 2507 (Thinking) | 43.33 | #260 (Qwen 3 235B A22B 2507) |
| 7 | Grok 3 Mini | 42.56 | #281 |
| 8 | DeepSeek V3.1 | 41.43 | #264 |
| 9 | DeepSeek R1 0528 | 41.37 | #222 |
| 10 | Gemini 2.5 Pro | 40.58 | #152 |
| 11 | Seed-OSS-36B-Instruct | 38.96 | #366 |
| 12 | Qwen 3 30B A3B 2507 (Thinking) | 37.33 | #376 (Qwen 3 30B A3B 2507) |
| 13 | Qwen 3 Next 80B A3B (Thinking) | 36.35 | #320 (Qwen 3 Next 80B A3B) |
| 14 | Claude Sonnet 4 (Thinking) | 30.51 | #199 (Claude Sonnet 4) |
| 15 | GLM-4.5 | 21.67 | #275 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=hardcorelogic-total · How It Works · Data refreshed daily, snapshot 2026-10-20.