reasoning benchmark
CruxEval-O
CruxEval-O is the output prediction task of the CRUXEval benchmark, designed to evaluate code reasoning, understanding, and execution capabilities. It consists of 800 Python functions (3-13 lines) where models must predict the output given a function and input. The benchmark tests fundamental code execution reasoning abilities and goes beyond simple code generation to assess deeper understanding of program behavior.
Updated Aug 7, 2026
Published models1
Registry coverage1
MetricScore
EvidenceB
CruxEval-O leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | MA | 51.3% | 100.0% | 1 | C |
CruxEval-O highlights
The top published results on this benchmark's own scale.
What is CruxEval-O?
Definition and scoring fields from the benchmark registry.
CruxEval-O is the output prediction task of the CRUXEval benchmark, designed to evaluate code reasoning, understanding, and execution capabilities. It consists of 800 Python functions (3-13 lines) where models must predict the output given a function and input. The benchmark tests fundamental code execution reasoning abilities and goes beyond simple code generation to assess deeper understanding of program behavior.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- CruxEval-O
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- cruxeval-o|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about CruxEval-O.
Which model scores highest on CruxEval-O?
Codestral-22B is currently ranked first with 51.3%.
What does CruxEval-O measure?
CruxEval-O is the output prediction task of the CRUXEval benchmark, designed to evaluate code reasoning, understanding, and execution capabilities. It consists of 800 Python functions (3-13 lines) where models must predict the output given a function and input. The benchmark tests fundamental code execution reasoning abilities and goes beyond simple code generation to assess deeper understanding of program behavior.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
1 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.