{"paper":{"title":"Qwen3Guard Technical Report","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Qwen3Guard provides multilingual guardrail models that output tri-class safety labels and monitor generation token by token.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"An Yang, Baosong Yang, Bowen Yu, Chen Cheng, Chenhan Yuan, Dayiheng Liu, Fei Huang, Haiquan Zhao, Jialong Tang, Jiandong Jiang, Jianwei Zhang, Jijie Xu, Jingren Zhou, Junyang Lin, Ming Yan, Minmin Sun, Pei Zhang, Pengjun Xie, Qiaoyu Tang, Qin Zhu, Rong Zhang, Shibin Wu, Shuo Zhang, Tao He, Tianyi Tang, Tingyu Xia, Wei Liao, Weizhou Shen, Wenbiao Yin, Wenmeng Zhou, Wenyuan Yu, Xiaobin Wang, Xiaodong Deng, Xiaodong Xu, Xiaomeng Hu, Xinyu Zhang, Yang Liu, Yeqiu Li, Yichang Zhang, Yi Zhang, Yong Jiang, Yu Wan, Yuxin Zhou","submitted_at":"2025-10-16T04:00:18Z","abstract_excerpt":"As large language models (LLMs) become more capable and widely used, ensuring the safety of their outputs is increasingly critical. Existing guardrail models, though useful in static evaluation settings, face two major limitations in real-world applications: (1) they typically output only binary \"safe/unsafe\" labels, which can be interpreted inconsistently across diverse safety policies, rendering them incapable of accommodating varying safety tolerances across domains; and (2) they require complete model outputs before performing safety checks, making them fundamentally incompatible with stre"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Evaluated across English, Chinese, and multilingual benchmarks, Qwen3Guard achieves state-of-the-art performance in both prompt and response safety classification.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The chosen benchmarks and tri-class/streaming formulations accurately reflect real-world safety needs and do not create new failure modes or policy inconsistencies when deployed at scale.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Qwen3Guard releases generative and streaming safety guard models in 0.6B/4B/8B sizes that deliver tri-class judgments and real-time token-level monitoring across 119 languages while claiming state-of-the-art benchmark results.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Qwen3Guard provides multilingual guardrail models that output tri-class safety labels and monitor generation token by token.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b33e561007a388d2fb77525d86612ef99e0693320c7776cf48e46481932c9cc9"},"source":{"id":"2510.14276","kind":"arxiv","version":1},"verdict":{"id":"eb60f1ec-7b5d-4643-91fb-198f8c8bf88d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T22:27:15.436293Z","strongest_claim":"Evaluated across English, Chinese, and multilingual benchmarks, Qwen3Guard achieves state-of-the-art performance in both prompt and response safety classification.","one_line_summary":"Qwen3Guard releases generative and streaming safety guard models in 0.6B/4B/8B sizes that deliver tri-class judgments and real-time token-level monitoring across 119 languages while claiming state-of-the-art benchmark results.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The chosen benchmarks and tri-class/streaming formulations accurately reflect real-world safety needs and do not create new failure modes or policy inconsistencies when deployed at scale.","pith_extraction_headline":"Qwen3Guard provides multilingual guardrail models that output tri-class safety labels and monitor generation token by token."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"a469441380dc5dbf4c048d107d3aab06f3f98f3c8aedd3682b55a5072fad3fd6"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}