{"paper":{"title":"An Industrial-Scale Insurance LLM Achieving Verifiable Domain Mastery and Hallucination Control without Competence Trade-offs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"INS-S1 reaches state-of-the-art insurance domain performance while preserving general capabilities and holding hallucinations to 0.6%.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jingjing Huo, Jun Li, Pan Liu, Qian Zhu, Wanqing Xu, Wenyan Yang, Xinnan Guo, Xuan Lin","submitted_at":"2026-03-15T16:13:37Z","abstract_excerpt":"Adapting Large Language Models (LLMs) to high-stakes vertical domains like insurance presents a significant challenge: scenarios demand strict adherence to complex regulations and business logic with zero tolerance for hallucinations. Existing approaches often suffer from a Competency Trade-off - sacrificing general intelligence for domain expertise - or rely heavily on RAG without intrinsic reasoning. To bridge this gap, we present INS-S1, an insurance-specific LLM family trained via a novel end-to-end alignment paradigm. Our approach features two methodological innovations: (1) A Verifiable "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"INS-S1 achieves SOTA performance on domain tasks, significantly outperforming DeepSeek-R1 and Gemini-2.5-Pro. Crucially, it maintains top-tier general capabilities and achieves a record-low 0.6% hallucination rate (HHEM).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the Progressive SFT-RL Curriculum Framework with dynamic data annealing and the mix of RLVR and RLAIF can enforce domain constraints and prevent catastrophic forgetting without any hidden trade-offs or post-training degradation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"INS-S1 is an insurance LLM that reaches SOTA domain performance with 0.6% hallucination rate and no loss in general intelligence via a new end-to-end alignment method and the INSEva benchmark.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"INS-S1 reaches state-of-the-art insurance domain performance while preserving general capabilities and holding hallucinations to 0.6%.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e035fb9b010ec9cddfc28cfe40c8aa4e325fb951acd08d9686172bb6ac8c8836"},"source":{"id":"2603.14463","kind":"arxiv","version":2},"verdict":{"id":"aabb3ae5-d2f3-4207-b879-51d468f21998","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T11:14:51.326694Z","strongest_claim":"INS-S1 achieves SOTA performance on domain tasks, significantly outperforming DeepSeek-R1 and Gemini-2.5-Pro. Crucially, it maintains top-tier general capabilities and achieves a record-low 0.6% hallucination rate (HHEM).","one_line_summary":"INS-S1 is an insurance LLM that reaches SOTA domain performance with 0.6% hallucination rate and no loss in general intelligence via a new end-to-end alignment method and the INSEva benchmark.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the Progressive SFT-RL Curriculum Framework with dynamic data annealing and the mix of RLVR and RLAIF can enforce domain constraints and prevent catastrophic forgetting without any hidden trade-offs or post-training degradation.","pith_extraction_headline":"INS-S1 reaches state-of-the-art insurance domain performance while preserving general capabilities and holding hallucinations to 0.6%."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2603.14463/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"edb81a9fb4d4d5996c27c0aac04b1bae512ec173648191da069d644dcd19f58d"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}