Intelligent Accounting and Financial Innovation from the Perspective of New Productivity: On-Balance Sheet and Disclosure Practices Based on Large Models

Large-language-model applications in intelligent accounting generate new technical assets whose recognition, measurement, and disclosure are difficult to handle under conventional static accounting frameworks. To address ambiguous asset-recognition standards, uncertain value measurement, and insufficient disclosure of model-related risks, this study constructs an LLM-based intelligent accounting asset-entry and disclosure framework. A three-layer asset identification model is developed to evaluate technical substance, expected economic benefit, and cost measurability. A dual-track measurement system combining cost approach and income approach is then established, with lifecycle cost tracking, excess-return estimation, amortization-period updating, and recoverable-amount testing. An intelligent disclosure template is further designed to report model architecture parameters, training-data sources, application performance, and risk-control indicators. Empirical analysis using 12 pilot enterprises over an 18-month period shows that the framework increases financial-statement readability from 62.3 to 85.7, improves audit efficiency by 42.3%, reduces substantive test samples by 42.4%, and increases investor decision relevance to 0.638. The study provides a structured information-system framework for technical asset measurement, model-risk disclosure, and data-driven financial decision support.