{"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&bench=MATH","json_url":"https://assignee.net/benchmarks/evidence.json?model=Qwen2&bench=MATH","model":"Qwen2","benchmark":"MATH","source_count":3,"source_coverage":{"record_count":3,"distinct_source_count":3,"coverage_level":"MODERATE","basis":"distinct public paper URLs or titles in this evidence cluster"},"source_profile":{"source_url_count":3,"missing_source_url_count":0,"domains":["arxiv.org"],"year_min":2024,"year_max":2025,"basis":"public source URLs, source titles, and reported publication years in this evidence cluster"},"reported_range":{"min_score_pct":43.0,"max_score_pct":67.4},"spread_pp":24.4,"severity":"HIGH","entries":[{"model":"Qwen2","benchmark":"MATH","score_pct":67.41,"source_title":"MINT-CoT: Enabling Interleaved Visual Tokens in Mathematical Chain-of-Thought Reasoning","source_url":"http://arxiv.org/abs/2506.05331v1","source_domain":"arxiv.org","year":2025},{"model":"Qwen2","benchmark":"MATH","score_pct":54.8,"source_title":"Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs","source_url":"http://arxiv.org/abs/2406.18629v1","source_domain":"arxiv.org","year":2024},{"model":"Qwen2","benchmark":"MATH","score_pct":43.04,"source_title":"CMM-Math: A Chinese Multimodal Math Dataset To Evaluate and Enhance the Mathematics Reasoning of Large Multimodal Models","source_url":"http://arxiv.org/abs/2409.02834v3","source_domain":"arxiv.org","year":2024}],"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."]}