Share your thoughts, 1 month free Claude Pro on usSee more

Code Reasoning on Code HumanEval+ LiveCodeBench v5

79.88HEval+ (Pass@1)

Qwen3-4B-Base (HEAL)

Updated 1mo ago

Evaluation Results

Method	Links
Qwen3-4B-Base (HEAL) 2026.04		79.88	26.95	53.42
Qwen3-4B-Base (Full-shot) 2026.04		78.05	23.95	51
Qwen3-4B-Base (Hybrid) 2026.04		74.39	24.55	49.47
Qwen3-4B-Base (Few-shot) 2026.04		71.95	22.75	47.35
Qwen3-4B-Base (Only-General) 2026.04		70.12	20.35	45.24
Qwen3-4B-Base 2026.04		66.46	16.77	41.62
Qwen3-1.7B-Base (HEAL) 2026.04		62.8	19.76	41.28
Qwen3-1.7B-Base (Full-shot) 2026.04		60.98	17.37	39.18
Qwen3-1.7B-Base (Only-General) 2026.04		58.54	10.18	34.36
Qwen3-1.7B-Base (Hybrid) 2026.04		57.93	13.77	35.85
Qwen3-1.7B-Base (Few-shot) 2026.04		56.71	12.57	34.64
Qwen3-1.7B-Base 2026.04		55.49	5.39	30.44
LLaMA3.2-3B-Instruct (HEAL) 2026.04		52.44	16.77	34.61
LLaMA3.2-3B-Instruct (Hybrid) 2026.04		51.83	15.57	33.7
LLaMA3.2-3B-Instruct (Full-shot) 2026.04		51.22	16.17	33.7
LLaMA3.2-3B-Instruct (Only-General) 2026.04		51.22	13.77	32.5
LLaMA3.2-3B-Instruct (Few-shot) 2026.04		50.61	12.57	31.59
LLaMA3.2-3B-Instruct 2026.04		48.17	11.38	29.78