{"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=GSM8K","json_url":"https://assignee.net/benchmarks/evidence.json?model=GPT-4&bench=GSM8K","model":"GPT-4","benchmark":"GSM8K","source_count":8,"source_coverage":{"record_count":8,"distinct_source_count":8,"coverage_level":"BROAD","basis":"distinct public paper URLs or titles in this evidence cluster"},"source_profile":{"source_url_count":8,"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":31.1,"max_score_pct":95.0},"spread_pp":63.9,"severity":"HIGH","entries":[{"model":"GPT-4","benchmark":"GSM8K","score_pct":95.0,"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":"GSM8K","score_pct":94.9,"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},{"model":"GPT-4","benchmark":"GSM8K","score_pct":92.7,"source_title":"Sample-Efficient Human Evaluation of Large Language Models via Maximum Discrepancy Competition","source_url":"http://arxiv.org/abs/2404.08008v2","source_domain":"arxiv.org","year":2024},{"model":"GPT-4","benchmark":"GSM8K","score_pct":84.2,"source_title":"Vendi-RAG: Adaptively Trading-Off Diversity And Quality Significantly Improves Retrieval Augmented Generation With LLMs","source_url":"http://arxiv.org/abs/2502.11228v2","source_domain":"arxiv.org","year":2025},{"model":"GPT-4","benchmark":"GSM8K","score_pct":80.0,"source_title":"Achieving >97% on GSM8K: Deeply Understanding the Problems Makes LLMs Better Solvers for Math Word Problems","source_url":"http://arxiv.org/abs/2404.14963v5","source_domain":"arxiv.org","year":2024},{"model":"GPT-4","benchmark":"GSM8K","score_pct":56.2,"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":"GSM8K","score_pct":40.0,"source_title":"MR-GSM8K: A Meta-Reasoning Benchmark for Large Language Model Evaluation","source_url":"http://arxiv.org/abs/2312.17080v4","source_domain":"arxiv.org","year":2023},{"model":"GPT-4","benchmark":"GSM8K","score_pct":31.1,"source_title":"GSM8K-V: Can Vision Language Models Solve Grade School Math Word Problems in Visual Contexts","source_url":"http://arxiv.org/abs/2509.25160v1","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."]}