Eval Connections One-Shot with Traps
Games · 2026-07-22
Eval Connections one-shot board on 20 canonical puzzles with reviewed trap sets. One scored submission of all four groups: 0/1/2 points for that many correct groups, 3 for all four, plus 2 for a correct trap claim or correct N/A. Maximum 100 points. The current implementation permits two structural-format repairs without correctness feedback. Kept separate from classic multi-turn play; per-run inference settings are not published in the aggregate table.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 94.0 |
| Grok 4.6 | 91.0 |
| Muse Spark 1.1 | 90.0 |
| GPT-6 Sol | 89.0 |
| Gemini 3.8 Flash | 88.0 |
| Claude Opus 5.5 | 86.0 |
| Gemini 3.1 Pro Preview | 86.0 |
| GPT-5.6 Terra | 86.0 |
| Claude Fable 5 | 85.0 |
| Grok 4.7 | 85.0 |
| Gemini 3 Flash Preview | 84.0 |
| Muse Spark 1.3 | 84.0 |
| GPT-5.6 Sol Pro | 84.0 |
| Kimi K3 | 84.0 |
| Fugu Ultra v1.0 | 84.0 |
| GPT-5.6 Sol | 83.0 |
| Claude Fable 5.1 | 82.0 |
| Grok 4.20 | 82.0 |
| Gemini 3.7 Flash | 80.0 |
| DeepSeek V4 Pro 0813 | 80.0 |
| GLM-5.3 Flash | 78.0 |
| Muse Spark 1.2 | 77.0 |
| GPT-5.6 Terra Pro | 77.0 |
| DeepSeek-V4-Flash-0731 | 77.0 |
| Grok 4.5 | 76.0 |
| Qwen3.8 Flash | 75.0 |
| Claude Opus 5 | 74.0 |
| GPT-5.6 Luna | 74.0 |
| GLM 5V Turbo | 74.0 |
| DeepSeek V4.1 Flash | 73.0 |
| GPT-5.6 Luna Pro | 72.0 |
| Sonnet 5 | 71.0 |
| Opus 4.5 | 70.0 |
| Gemini 3.6 Flash | 69.0 |
| MiMo-V2.6-Flash | 69.0 |
| Gemini 3.5 Flash-Lite | 67.0 |
| Gemini 3.5 Flash | 67.0 |
| Inkling | 67.0 |
| Qwen3.8 2.4T A95B | 66.0 |
| Opus 4.7 | 65.0 |