Pencil Puzzle Bench - Direct Ask
Games · 2026-03-02
Pencil Puzzle Bench's Direct Ask setting presents 300 pencil puzzles directly to models. Scores report the percentage solved correctly.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 50.0 |
| Opus 4.6 | 27.3 |
| GPT-5.2 | 27.0 |
| Gemini 3.1 Pro Preview | 20.0 |
| GLM-5.2 | 16.0 |
| Gemini 3.5 Flash | 13.0 |
| Kimi K2.6 | 11.7 |
| Sonnet 4.6 | 10.3 |
| GPT-5.2 Pro | 9.7 |
| GPT-5.1 | 7.7 |
| Qwen3.7-Max | 7.3 |
| Kimi K2.7 Code | 6.7 |
| Opus 4.5 | 6.0 |
| GPT-5 | 6.0 |
| Kimi K2.5 | 6.0 |
| Grok 4.1 Fast | 5.7 |
| Qwen3.7-Plus | 5.3 |
| Gemini 3 Flash Preview | 4.7 |
| Gemini 3 Pro Preview | 4.3 |
| DeepSeek-V4-Pro | 4.0 |
| Grok 4.3 | 3.3 |
| MiniMax M3 | 3.3 |
| Nemotron 3 Ultra 550B A55B | 3.3 |
| o3 | 3.0 |
| Sonnet 4.5 | 2.3 |
| DeepSeek-V3.2-Speciale | 2.3 |
| DeepSeek-V3.2 | 2.0 |
| Kimi K2 Thinking | 1.3 |
| MiMo-V2-Pro | 1.0 |
| GLM-5 | 0.7 |
| MiniMax M2.5 | 0.7 |
| MiniMax M2.7 | 0.7 |
| O1 | 0.7 |
| Qwen3.5 397B A17B | 0.7 |
| Gemini 2.5 Pro | 0.3 |
| Gemma 4 31B IT | 0.3 |
| GLM-4.7 | 0.3 |
| gpt-oss-120b | 0.3 |
| Grok 4.20 | 0.3 |
| Grok Code Fast 1 | 0.3 |