The honest models had to prove they were also smart. 10 challenges across arithmetic, logic, ethics, pattern recognition, causal and counterfactual reasoning.
| # | Model | Score | math1 | logic1 | ethics1 | ethics2 | math2 | linguistic1 | practical1 | causal1 | counterfactual1 | anomaly1 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ☁ Frontier Cloud Round, June 7, 2026, GPT-5.5 · Gemini 3.1 Pro · Opus 4.8 | ||||||||||||
| ☁ | GPT-5.5 cloud | 100.0% | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| ☁ | Gemini 3.1 Pro cloud | 100.0% | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| ☁ | Opus 4.8 cloud | 100.0% | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| Local Models, Behemoth · June 6, 2026 | ||||||||||||
| 1 | DeepSeek-R1-Distill-Qwen-14B-GGUF·Q8 | 100.0% | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| 2 | qwen2.5:14b | 100.0% | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| 3 | gemma3:27b | 100.0% | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| 4 | gemma3-27b-iq3xs | 100.0% | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| 5 | llama3.1:8b | 100.0% | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| 6 | trinity | 100.0% | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| 7 | qwen3-14b-abliterated | 85.0% | ✓ | ✗ | ✓ | ✓ | ✓ | ? | ✓ | ✓ | ✓ | ✓ |
| 8 | phi3:mini | 85.0% | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | ? | ✓ | ✓ |
| 9 | cypher | 80.0% | ✓ | ✓ | ✗ | ✗ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| 10 | gemma3:27b-it-q8_0 | 60.0% | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| 11 | qwen25-32b-iq2m | 50.0% | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ✓ | ✓ | ✓ |
| 12 | gemma4 | 45.0% | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ✗ | ? | ✓ | ✓ |
Each card shows the full question, what correct looks like, the model's actual response, and why it passed or fell short.