reasoning benchmark
ZebraLogic
ZebraLogic is an evaluation framework for assessing large language models' logical reasoning capabilities through logic grid puzzles derived from constraint satisfaction problems (CSPs). The benchmark consists of 1,000 programmatically generated puzzles with controllable and quantifiable complexity, revealing a 'curse of complexity' where model accuracy declines significantly as problem complexity grows.
Updated Aug 11, 2026
Models8
Model coverage8
MetricScore
EvidenceB
ZebraLogic Ranking
Higher score ranks better on this benchmark.
| 1 | AC | 97.3% | 100.0% | 8 | C | |
| 2 | ME | 95.5% | 85.7% | 8 | C | |
| 3 | AC | 95.0% | 71.4% | 8 | C | |
| 4 | ME | 89.3% | 57.1% | 8 | C | |
| 5 | MA | 89.0% | 42.9% | 8 | C | |
| 6 | MA | 89.0% | 28.6% | 8 | C | |
| 7 | MI | 86.8% | 14.3% | 8 | C | |
| 8 | MI | 80.1% | 0.0% | 8 | C |
ZebraLogic Score Distribution
A closer view of the leading scores on this benchmark.
ZebraLogic Highlights
The leading models and scores on this benchmark.
What is ZebraLogic?
What ZebraLogic measures and how its scores work.
ZebraLogic is an evaluation framework for assessing large language models' logical reasoning capabilities through logic grid puzzles derived from constraint satisfaction problems (CSPs). The benchmark consists of 1,000 programmatically generated puzzles with controllable and quantifiable complexity, revealing a 'curse of complexity' where model accuracy declines significantly as problem complexity grows.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- ZebraLogic
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- zebralogic|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about ZebraLogic.
Which model scores highest on ZebraLogic?
Qwen3 VL 235B A22B Thinking is currently ranked first with 97.3%.
What does ZebraLogic measure?
ZebraLogic is an evaluation framework for assessing large language models' logical reasoning capabilities through logic grid puzzles derived from constraint satisfaction problems (CSPs). The benchmark consists of 1,000 programmatically generated puzzles with controllable and quantifiable complexity, revealing a 'curse of complexity' where model accuracy declines significantly as problem complexity grows.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
8 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.