{"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-4o&bench=MATH","json_url":"https://assignee.net/benchmarks/evidence.json?model=GPT-4o&bench=MATH","model":"GPT-4o","benchmark":"MATH","source_count":6,"source_coverage":{"record_count":6,"distinct_source_count":6,"coverage_level":"BROAD","basis":"distinct public paper URLs or titles in this evidence cluster"},"source_profile":{"source_url_count":6,"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":29.0,"max_score_pct":95.0},"spread_pp":66.0,"severity":"HIGH","entries":[{"model":"GPT-4o","benchmark":"MATH","score_pct":95.0,"source_title":"Benchmarking LLMs' Mathematical Reasoning with Unseen Random Variables Questions","source_url":"http://arxiv.org/abs/2501.11790v4","source_domain":"arxiv.org","year":2025},{"model":"GPT-4o","benchmark":"MATH","score_pct":85.9,"source_title":"AceMath: Advancing Frontier Math Reasoning with Post-Training and Reward Modeling","source_url":"http://arxiv.org/abs/2412.15084v2","source_domain":"arxiv.org","year":2024},{"model":"GPT-4o","benchmark":"MATH","score_pct":76.7,"source_title":"Safe: Enhancing Mathematical Reasoning in Large Language Models via Retrospective Step-aware Formal Verification","source_url":"http://arxiv.org/abs/2506.04592v1","source_domain":"arxiv.org","year":2025},{"model":"GPT-4o","benchmark":"MATH","score_pct":73.4,"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-4o","benchmark":"MATH","score_pct":72.67,"source_title":"Beyond Meta-Reasoning: Metacognitive Consolidation for Self-Improving LLM Reasoning","source_url":"http://arxiv.org/abs/2604.17399v1","source_domain":"arxiv.org","year":2026},{"model":"GPT-4o","benchmark":"MATH","score_pct":29.02,"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."]}