An Industrial-Scale Insurance LLM Achieving Verifiable Domain Mastery and Hallucination Control without Competence Trade-offs

The article presents INS-S1, a large language model specifically designed for the insurance domain, utilizing an innovative end-to-end alignment paradigm that includes a Verifiable Data Synthesis System and a Progressive SFT-RL Curriculum Framework. This model achieves state-of-the-art performance on domain-specific tasks, outperforming competitors like DeepSeek-R1 and Gemini-2.5-Pro, while maintaining a low hallucination rate of 0.6%. The introduction of INSEva, a comprehensive benchmark with over 39,000 samples, provides a valuable resource for evaluating insurance LLMs, indicating that rigorous domain specialization can be achieved without sacrificing general intelligence.