Pencil Puzzle Bench - Mashu — leaderboard

Metric: Direct-ask Success Rate (%). Source: ppbench.com. 51 models tracked.

Top models

#ModelScore
1Claude Opus 4.6 (Thinking)60
2GPT-5.2 (xHigh)60
3GPT-5.2 (High)33.3
4Gemini 3.1 Pro (Preview)20
5GPT-5.2 (Medium)13.3
6DeepSeek V3.26.7
7Gemini 3 Pro6.7
8Grok 4.1 Fast6.7
9Claude Opus 4.5 (Thinking)6.7
10GPT-5.1 (Medium)6.7
11GPT-5.2 Pro6.7
12Gemini 3 Flash (High)6.7
13Claude Sonnet 4.6 (Thinking)6.7
14Gemini 2.5 Pro0
15GPT-4o0

Interactive version: theaggregate.ai/benchmark?slug=pencil-puzzle-bench-mashu · How the rankings work · Data refreshed daily, snapshot 2026-07-22.