{"paper":{"title":"CMMLU: Measuring massive multitask language understanding in Chinese","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Most large language models score below 50 percent on a new Chinese multitask understanding benchmark.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Fajri Koto, Hai Zhao, Haonan Li, Nan Duan, Timothy Baldwin, Yeyun Gong, Yifei Yang, Yixuan Zhang","submitted_at":"2023-06-15T15:49:51Z","abstract_excerpt":"As the capabilities of large language models (LLMs) continue to advance, evaluating their performance becomes increasingly crucial and challenging. This paper aims to bridge this gap by introducing CMMLU, a comprehensive Chinese benchmark that covers various subjects, including natural science, social sciences, engineering, and humanities. We conduct a thorough evaluation of 18 advanced multilingual- and Chinese-oriented LLMs, assessing their performance across different subjects and settings. The results reveal that most existing LLMs struggle to achieve an average accuracy of 50%, even when "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"most existing LLMs struggle to achieve an average accuracy of 50%, even when provided with in-context examples and chain-of-thought prompts, whereas the random baseline stands at 25%.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The questions in CMMLU accurately represent the knowledge and reasoning demands of real Chinese-language tasks across the covered subjects.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CMMLU benchmark shows most advanced LLMs score below 50% accuracy on Chinese multitask understanding, well above the 25% random baseline but revealing major room for improvement.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Most large language models score below 50 percent on a new Chinese multitask understanding benchmark.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"afd9388db84a4af3ea61c6f5c90367cea2a2ecbba52536f98631f829755bf1fe"},"source":{"id":"2306.09212","kind":"arxiv","version":2},"verdict":{"id":"7e67a584-1eaf-49f9-b182-dc341b1f0120","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T20:58:16.649845Z","strongest_claim":"most existing LLMs struggle to achieve an average accuracy of 50%, even when provided with in-context examples and chain-of-thought prompts, whereas the random baseline stands at 25%.","one_line_summary":"CMMLU benchmark shows most advanced LLMs score below 50% accuracy on Chinese multitask understanding, well above the 25% random baseline but revealing major room for improvement.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The questions in CMMLU accurately represent the knowledge and reasoning demands of real Chinese-language tasks across the covered subjects.","pith_extraction_headline":"Most large language models score below 50 percent on a new Chinese multitask understanding benchmark."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"9600d38a99db9286fc9a63dad72fb1478b226bdd6e157f3b3acdfdb5cd52bdde"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}