Code Benchmark

HumanEval+

HumanEval+ is the EvalPlus extension of HumanEval with 80x more generated unit tests for correctness checking. Score recorded from the Bonsai 2 27B GGUF card's Full Per-Benchmark Results (thinking mode, EvalScope + vLLM on NVIDIA H100).

Category

Code

Score type

percent

Scale

100%

higher is better

Models tested

3

What is HumanEval+?

HumanEval+ is a benchmark for code generation and code reasoning. The official description says: "HumanEval+ is the EvalPlus extension of HumanEval with 80x more generated unit tests for correctness checking". Below we describe in simple words what benchmarks of this type generally examine.

What does it test?

HumanEval+ measures code generation and code reasoning. In detail, it checks:

  • writing programs that must pass hidden unit tests
  • predicting what a given piece of code outputs (true code comprehension)
  • handling edge cases, not just the happy path
  • producing runnable, correct solutions from short specifications

How to read the score

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

Official methodology note: HumanEval+ is the EvalPlus extension of HumanEval with 80x more generated unit tests for correctness checking. Score recorded from the Bonsai 2 27B GGUF card's Full Per-Benchmark Results (thinking mode, EvalScope + vLLM on NVIDIA H100).

Frequently asked questions

What is the HumanEval+ benchmark?

HumanEval+ is a benchmark for code generation and code reasoning. The official description says: 'HumanEval+ is the EvalPlus extension of HumanEval with 80x more generated unit tests for correctness checking'. Below we describe in simple words what benchmarks of this type generally examine.

What does HumanEval+ test?

HumanEval+ mainly measures code generation and code reasoning. Concretely, it checks: writing programs that must pass hidden unit tests; predicting what a given piece of code outputs (true code comprehension); handling edge cases, not just the happy path.

How is the HumanEval+ score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get HumanEval+ data and scores?

The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/humaneval-plus/scores) or pull it directly from the Open LLM Radar API with any HTTP client.

Model scores 3 models tested

Reset
# Model Score Date Source Verified
1 Bonsai 2 27B — GGUF
Prism ML
95.12%
17.09.2026 Report Verified
2 Kolibri-1
Aleph Alpha
92.7%
03.10.2026 Report Verified
3 ERNIE-5.5-Flash-Base
Baidu
82.3%
— HF model card baidu/ERNIE-5.5-Flash-Base (Pass@1, 1-shot, Unified Pipeline) Unverified

What now?

You've seen how models score on HumanEval+ — here's what to do next.