In a landmark technological advance that reshapes global climate action infrastructure, China publicly launched a groundbreaking generative artificial intelligence large model dedicated to carbon emission accounting in Shanghai on Wednesday. This launch marks the first full-spectrum carbon accounting system in the world that integrates measurement across production-side emissions, consumption-side emissions, and natural carbon sources, according to its developer, the Shanghai Advanced Research Institute under the Chinese Academy of Sciences (CAS).
Carbon emission accounting serves as a non-negotiable foundation for global climate policy compliance, a core underpinning for international carbon pricing mechanisms, and a mandatory prerequisite for nations working to meet their peak carbon and carbon neutrality commitments. For decades, the field has been held back by persistent bottlenecks: steep knowledge barriers for practitioners, convoluted and time-consuming data processing workflows, long analysis timelines, and low spatial and temporal resolution in results. The new large model is specifically engineered to overcome these limitations, leveraging generative AI to rewrite the standard operating paradigm of carbon accounting.
Built upon CAS’s existing foundational scientific model ScienceOne, the new carbon accounting model rests on three core technical pillars. First, it incorporates eight independently owned proprietary datasets that support high-frequency data updates and seamless cross-source data fusion. Second, it relies on a home-grown methodological framework powered by a large language model-based multi-agent collaboration system, which drastically improves the accuracy of accounting results. Third, it operates on a hybrid computing cluster that optimizes resource allocation across internal institutional servers and external high-performance computing centers.
Currently, the model’s open service interface hosts a 32-billion-parameter vertical-domain large language model paired with an intelligent emissions database, offering access through both conversational user interfaces and open programming interfaces for developers and researchers. Five functionally distinct specialized intelligent agents have been integrated into the system, each tailored to handle specific core tasks: digital simulation and process optimization for industrial systems, accounting for carbon transfer embedded in cross-border trade, full product life cycle assessment, natural carbon source accounting, and systematic uncertainty analysis of results.
Of particular note is the life cycle assessment agent, which can autonomously complete the entire end-to-end workflow of product carbon footprint accounting — from defining assessment goals and scope, compiling emissions inventories, conducting formal accounting, to interpreting final results — fully automating a process that previously required extensive manual input from specialized experts.
Building on the model’s high-resolution calculation capabilities, research teams have already completed an initial high-precision national-level carbon holographic map. Using 2022 emissions data as a test case, the model’s new, scientifically more equitable accounting framework produced adjusted emission figures for major economies that differed significantly from traditional production-side calculations published by the Intergovernmental Panel on Climate Change (IPCC): China’s total emissions were adjusted downward by 17.7%, while the United States’ emissions were adjusted upward by 15.2% and Japan’s by 7.2%.
The model’s analysis also uncovered a key systemic bias in current international carbon policy: the default emission factors used in the European Union’s Carbon Border Adjustment Mechanism (CBAM) systematically overestimate the carbon intensity of Chinese manufactured goods, a finding that underscores the urgent need for more accurate accounting and the adoption of localized, region-specific emission factors for trade policy.
Beyond identifying structural gaps in global carbon governance, the model also quantifies the global climate benefits of China’s green technology exports. For example, it calculated that Chinese-produced wind turbines and photovoltaic products exported in 2024 generated roughly 2 million tonnes of carbon emissions during their domestic manufacturing phase, but will deliver an estimated 350 million tonnes of cumulative carbon emission reductions during their operational lifetime around the world.
For China, the new model provides critical technical support for compiling national greenhouse gas inventories, developing the national carbon trading market, driving the green transition of high-emission key industries, and formulating evidence-based responses to international carbon-related trade policies. On the global stage, the breakthrough offers Chinese technical expertise to international efforts to build a fairer, more scientifically rigorous global system for carbon accounting and climate responsibility allocation, strengthening China’s technological influence in global climate governance.