{"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=GPT-4&bench=MATH","json_url":"https://assignee.net/benchmarks/evidence.json?model=GPT-4&bench=MATH","model":"GPT-4","benchmark":"MATH","source_count":5,"source_coverage":{"record_count":5,"distinct_source_count":5,"coverage_level":"BROAD","basis":"distinct public paper URLs or titles in this evidence cluster"},"source_profile":{"source_url_count":5,"missing_source_url_count":0,"domains":["arxiv.org"],"year_min":2023,"year_max":2026,"basis":"public source URLs, source titles, and reported publication years in this evidence cluster"},"reported_range":{"min_score_pct":29.8,"max_score_pct":69.7},"spread_pp":39.9,"severity":"HIGH","entries":[{"model":"GPT-4","benchmark":"MATH","score_pct":69.69,"source_title":"Solving Challenging Math Word Problems Using GPT-4 Code Interpreter with Code-based Self-Verification","source_url":"http://arxiv.org/abs/2308.07921v1","source_domain":"arxiv.org","year":2023},{"model":"GPT-4","benchmark":"MATH","score_pct":60.1,"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":"GPT-4","benchmark":"MATH","score_pct":48.1,"source_title":"Math-Shepherd: Verify and Reinforce LLMs Step-by-step without Human Annotations","source_url":"http://arxiv.org/abs/2312.08935v3","source_domain":"arxiv.org","year":2023},{"model":"GPT-4","benchmark":"MATH","score_pct":42.5,"source_title":"Chain-of-Thought Hub: A Continuous Effort to Measure Large Language Models' Reasoning Performance","source_url":"http://arxiv.org/abs/2305.17306v1","source_domain":"arxiv.org","year":2023},{"model":"GPT-4","benchmark":"MATH","score_pct":29.8,"source_title":"Is Mathematical Problem-Solving Expertise in Large Language Models Associated with Assessment Performance?","source_url":"http://arxiv.org/abs/2603.25633v1","source_domain":"arxiv.org","year":2026}],"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."]}