{"schema":"https://assignee.net/schemas/benchmark-evidence-v1","schema_version":"1.0","contract_version":"benchmark-evidence-v1.0","contract_updated":"2026-06-01","schema_documentation":"https://assignee.net/schemas","changelog_url":"https://assignee.net/changelog","publisher":{"name":"Assignee Research","url":"https://assignee.net"},"html_url":"https://assignee.net/benchmarks/evidence?model=Qwen2.5&bench=HumanEval","json_url":"https://assignee.net/benchmarks/evidence.json?model=Qwen2.5&bench=HumanEval","model":"Qwen2.5","benchmark":"HumanEval","source_count":7,"source_coverage":{"record_count":7,"distinct_source_count":7,"coverage_level":"BROAD","basis":"distinct public paper URLs or titles in this evidence cluster"},"source_profile":{"source_url_count":7,"missing_source_url_count":0,"domains":["arxiv.org"],"year_min":2024,"year_max":2026,"basis":"public source URLs, source titles, and reported publication years in this evidence cluster"},"reported_range":{"min_score_pct":32.6,"max_score_pct":96.3},"spread_pp":63.7,"severity":"HIGH","entries":[{"model":"Qwen2.5","benchmark":"HumanEval","score_pct":96.34,"source_title":"ReflexiCoder: Teaching Large Language Models to Self-Reflect on Generated Code and Self-Correct It via Reinforcement Learning","source_url":"https://arxiv.org/abs/2603.05863","source_domain":"arxiv.org","year":2026},{"model":"Qwen2.5","benchmark":"HumanEval","score_pct":82.2,"source_title":"FeedbackEval: A Benchmark for Evaluating Large Language Models in Feedback-Driven Code Repair Tasks","source_url":"https://arxiv.org/abs/2504.06939","source_domain":"arxiv.org","year":2025},{"model":"Qwen2.5","benchmark":"HumanEval","score_pct":79.6,"source_title":"FeedbackEval: A Benchmark for Evaluating Large Language Models in Feedback-Driven Code Repair Tasks","source_url":"http://arxiv.org/abs/2504.06939v2","source_domain":"arxiv.org","year":2025},{"model":"Qwen2.5","benchmark":"HumanEval","score_pct":59.6,"source_title":"HumanEval Pro and MBPP Pro: Evaluating Large Language Models on Self-invoking Code Generation","source_url":"http://arxiv.org/abs/2412.21199v2","source_domain":"arxiv.org","year":2024},{"model":"Qwen2.5","benchmark":"HumanEval","score_pct":59.1,"source_title":"Qwen2.5 Technical Report","source_url":"http://arxiv.org/abs/2412.15115v2","source_domain":"arxiv.org","year":2024},{"model":"Qwen2.5","benchmark":"HumanEval","score_pct":41.0,"source_title":"Assessing Small Language Models for Code Generation: An Empirical Study with Benchmarks","source_url":"http://arxiv.org/abs/2507.03160v4","source_domain":"arxiv.org","year":2025},{"model":"Qwen2.5","benchmark":"HumanEval","score_pct":32.6,"source_title":"LLaDA-MoE: A Sparse MoE Diffusion Language Model","source_url":"http://arxiv.org/abs/2509.24389v1","source_domain":"arxiv.org","year":2025}],"interpretation":"This record groups score claims extracted from papers for the same model and benchmark label. A nonzero spread means the public literature reports different values for this cluster.","limitations":["Differences are not automatically errors.","Reported values may differ because of prompts, dataset versions, evaluation protocols, scoring rules, preprocessing, fine-tuning, or reporting conventions.","Source papers remain authoritative for their own claims."]}