reasoning benchmark
HumanEval+
Enhanced version of HumanEval that extends the original test cases by 80x using EvalPlus framework for rigorous evaluation of LLM-synthesized code functional correctness, detecting previously undetected wrong code
Updated Aug 11, 2026
Models10
Model coverage10
MetricScore
EvidenceB
HumanEval+ Ranking
Higher score ranks better on this benchmark.
| 1 | MI | 92.9% | 100.0% | 10 | C | |
| 2 | MI | 92.3% | 88.9% | 10 | C | |
| 3 | IB | 86.1% | 77.8% | 10 | C | |
| 4 | IB | 86.1% | 66.7% | 10 | C | |
| 5 | MI | 82.8% | 55.6% | 10 | C | |
| 6 | IB | 78.3% | 44.4% | 10 | C | |
| 7 | XI | 75.6% | 33.3% | 10 | C | |
| 8 | AC | 52.4% | 22.2% | 10 | C | |
| 9 | AC | 51.2% | 11.1% | 10 | C | |
| 10 | BA | 25.0% | 0.0% | 10 | C |
HumanEval+ Score Distribution
A closer view of the leading scores on this benchmark.
HumanEval+ Highlights
The leading models and scores on this benchmark.
What is HumanEval+?
What HumanEval+ measures and how its scores work.
Enhanced version of HumanEval that extends the original test cases by 80x using EvalPlus framework for rigorous evaluation of LLM-synthesized code functional correctness, detecting previously undetected wrong code
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- HumanEval+
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- humaneval+|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about HumanEval+.
Which model scores highest on HumanEval+?
Phi 4 Reasoning is currently ranked first with 92.9%.
What does HumanEval+ measure?
Enhanced version of HumanEval that extends the original test cases by 80x using EvalPlus framework for rigorous evaluation of LLM-synthesized code functional correctness, detecting previously undetected wrong code
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
10 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.