// BENCHMARK

ZebraLogic benchmark

AI model leaderboard for the ZebraLogic benchmark. Compare how large language models score on ZebraLogic, see the full ranking, and understand what this AI benchmark measures. GPT-5 currently leads with 98%. 1,000 logic-grid puzzles ("zebra puzzles") covering 6 difficulty tiers.

Leaderboard

# Model Organization Score Variant Source
#1 GPT-5 OpenAI 98% thinking-high-cited-ring-1t official ↗
#2 Qwen3-235B-A22B-Thinking (Jul 2025) Qwen 97.03% thinking-cited-ring-1t official ↗
#3 DeepSeek-V3.1-Terminus DeepSeek 96.33% thinking-cited-ring-1t official ↗
#4 Ring-1T Ant Group 95.15% thinking official ↗
#5 gemini-2.5-pro Google DeepMind 92.4% cited-ring-1t official ↗
#6 Ling-1T Ant Group 90.8% pass@1 official ↗
#7 Qwen3-30B-A3B Qwen 89.5% thinking official ↗
#8 LongCat-Flash Meituan Inc 89.3% official ↗
#9 Qwen3-32B Qwen 88.8% thinking official ↗
#10 Qwen3-14B Qwen 88.5% thinking official ↗
#11 MiniMax-M1-80k MiniMax 86.8% official ↗
#12 Kimi K2 Moonshot 85.5% instruct-0905-cited-ling-1t official ↗
#13 Qwen3-8B Qwen 84.8% thinking official ↗
#14 DeepSeek-V3.1-Terminus DeepSeek 81.6% non-thinking-cited-ling-1t official ↗
#15 Qwen3-4B Qwen 81% thinking official ↗
#16 Qwen3-235B-A22B Qwen 80.3% thinking official ↗
#17 MiniMax-M1-40k MiniMax 80.1% official ↗
#18 gemini-2.5-pro Google DeepMind 70.2% lowthink-cited-ling-1t official ↗
#19 Qwen3-1.7B Qwen 63.2% thinking official ↗
#20 GPT-5 OpenAI 57.3% main-cited-ling-1t official ↗
#21 Qwen3-235B-A22B Qwen 37.7% non-thinking official ↗
#22 Qwen3-4B Qwen 35.2% non-thinking official ↗
#23 Qwen3-30B-A3B Qwen 33.2% non-thinking official ↗
#24 Qwen3-14B Qwen 33% non-thinking official ↗
#25 Qwen3-0.6B Qwen 30.3% thinking official ↗
#26 Qwen3-32B Qwen 29.2% non-thinking official ↗
#27 Qwen3-8B Qwen 26.7% non-thinking official ↗
#28 Qwen3-1.7B Qwen 12.8% non-thinking official ↗
#29 Qwen3-0.6B Qwen 4.2% non-thinking official ↗

Frequently asked questions about ZebraLogic

What is the ZebraLogic benchmark?

1,000 logic-grid puzzles ("zebra puzzles") covering 6 difficulty tiers.

How is the ZebraLogic benchmark scored?

ZebraLogic is scored using the accuracy (%) metric, where a higher score is better. The maximum achievable score is 100.000. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.

Which AI model scores highest on ZebraLogic?

As of the latest reported scores on GenAIList, GPT-5 achieves the highest result on ZebraLogic with a score of 98%.

Is a higher ZebraLogic score better?

Yes. On ZebraLogic a higher score indicates better performance, so models near the top of the leaderboard are the strongest.